projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
untangling do_lookup() - switch to calling __lookup_hash()
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / namei.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
630d9c47 18#include <linux/export.h>
1da177e4
LT		 19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
1da177e4 22#include <linux/pagemap.h>
0eeca283 23#include <linux/fsnotify.h>
1da177e4
LT		 24#include <linux/personality.h>
25#include <linux/security.h>
6146f0d5 26#include <linux/ima.h>
1da177e4
LT		 27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
16f7e0fe 30#include <linux/capability.h>
834f2a4a 31#include <linux/file.h>
5590ff0d 32#include <linux/fcntl.h>
08ce5f16 33#include <linux/device_cgroup.h>
5ad4e53b 34#include <linux/fs_struct.h>
e77819e5 35#include <linux/posix_acl.h>
1da177e4
LT		 36#include <asm/uaccess.h>
37
e81e3f4d 38#include "internal.h"
c7105365 39#include "mount.h"
e81e3f4d 40
1da177e4
LT		 41/* [Feb-1997 T. Schoebel-Theuer]
42 * Fundamental changes in the pathname lookup mechanisms (namei)
43 * were necessary because of omirr. The reason is that omirr needs
44 * to know the _real_ pathname, not the user-supplied one, in case
45 * of symlinks (and also when transname replacements occur).
46 *
47 * The new code replaces the old recursive symlink resolution with
48 * an iterative one (in case of non-nested symlink chains). It does
49 * this with calls to <fs>_follow_link().
50 * As a side effect, dir_namei(), _namei() and follow_link() are now
51 * replaced with a single function lookup_dentry() that can handle all
52 * the special cases of the former code.
53 *
54 * With the new dcache, the pathname is stored at each inode, at least as
55 * long as the refcount of the inode is positive. As a side effect, the
56 * size of the dcache depends on the inode cache and thus is dynamic.
57 *
58 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
59 * resolution to correspond with current state of the code.
60 *
61 * Note that the symlink resolution is not *completely* iterative.
62 * There is still a significant amount of tail- and mid- recursion in
63 * the algorithm. Also, note that <fs>_readlink() is not used in
64 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
65 * may return different results than <fs>_follow_link(). Many virtual
66 * filesystems (including /proc) exhibit this behavior.
67 */
68
69/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
70 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
71 * and the name already exists in form of a symlink, try to create the new
72 * name indicated by the symlink. The old code always complained that the
73 * name already exists, due to not following the symlink even if its target
74 * is nonexistent. The new semantics affects also mknod() and link() when
25985edc 75 * the name is a symlink pointing to a non-existent name.
1da177e4
LT		 76 *
77 * I don't know which semantics is the right one, since I have no access
78 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
79 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
80 * "old" one. Personally, I think the new semantics is much more logical.
81 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
82 * file does succeed in both HP-UX and SunOs, but not in Solaris
83 * and in the old Linux semantics.
84 */
85
86/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
87 * semantics. See the comments in "open_namei" and "do_link" below.
88 *
89 * [10-Sep-98 Alan Modra] Another symlink change.
90 */
91
92/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
93 * inside the path - always follow.
94 * in the last component in creation/removal/renaming - never follow.
95 * if LOOKUP_FOLLOW passed - follow.
96 * if the pathname has trailing slashes - follow.
97 * otherwise - don't follow.
98 * (applied in that order).
99 *
100 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
101 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
102 * During the 2.4 we need to fix the userland stuff depending on it -
103 * hopefully we will be able to get rid of that wart in 2.5. So far only
104 * XEmacs seems to be relying on it...
105 */
106/*
107 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 108 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 109 * any extra contention...
110 */
111
112/* In order to reduce some races, while at the same time doing additional
113 * checking and hopefully speeding things up, we copy filenames to the
114 * kernel data space before using them..
115 *
116 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
117 * PATH_MAX includes the nul terminator --RR.
118 */
858119e1 119static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 120{
121 int retval;
122 unsigned long len = PATH_MAX;
123
124 if (!segment_eq(get_fs(), KERNEL_DS)) {
125 if ((unsigned long) filename >= TASK_SIZE)
126 return -EFAULT;
127 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
128 len = TASK_SIZE - (unsigned long) filename;
129 }
130
131 retval = strncpy_from_user(page, filename, len);
132 if (retval > 0) {
133 if (retval < len)
134 return 0;
135 return -ENAMETOOLONG;
136 } else if (!retval)
137 retval = -ENOENT;
138 return retval;
139}
140
1fa1e7f6 141static char *getname_flags(const char __user *filename, int flags, int *empty)
1da177e4 142{
4043cde8
EP		 143 char *result = __getname();
144 int retval;
145
146 if (!result)
147 return ERR_PTR(-ENOMEM);
148
149 retval = do_getname(filename, result);
150 if (retval < 0) {
151 if (retval == -ENOENT && empty)
152 *empty = 1;
153 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
154 __putname(result);
155 return ERR_PTR(retval);
1da177e4
LT		 156 }
157 }
158 audit_getname(result);
159 return result;
160}
161
f52e0c11
AV		 162char *getname(const char __user * filename)
163{
f7493e5d 164 return getname_flags(filename, 0, NULL);
f52e0c11
AV		 165}
166
1da177e4
LT		 167#ifdef CONFIG_AUDITSYSCALL
168void putname(const char *name)
169{
5ac3a9c2 170 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 171 audit_putname(name);
172 else
173 __putname(name);
174}
175EXPORT_SYMBOL(putname);
176#endif
177
e77819e5
LT		 178static int check_acl(struct inode *inode, int mask)
179{
84635d68 180#ifdef CONFIG_FS_POSIX_ACL
e77819e5
LT		 181 struct posix_acl *acl;
182
e77819e5 183 if (mask & MAY_NOT_BLOCK) {
3567866b
AV		 184 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
185 if (!acl)
e77819e5 186 return -EAGAIN;
3567866b
AV		 187 /* no ->get_acl() calls in RCU mode... */
188 if (acl == ACL_NOT_CACHED)
189 return -ECHILD;
206b1d09 190 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
e77819e5
LT		 191 }
192
193 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
194
195 /*
4e34e719
CH		 196 * A filesystem can force a ACL callback by just never filling the
197 * ACL cache. But normally you'd fill the cache either at inode
198 * instantiation time, or on the first ->get_acl call.
e77819e5 199 *
4e34e719
CH		 200 * If the filesystem doesn't have a get_acl() function at all, we'll
201 * just create the negative cache entry.
e77819e5
LT		 202 */
203 if (acl == ACL_NOT_CACHED) {
4e34e719
CH		 204 if (inode->i_op->get_acl) {
205 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
206 if (IS_ERR(acl))
207 return PTR_ERR(acl);
208 } else {
209 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
210 return -EAGAIN;
211 }
e77819e5
LT		 212 }
213
214 if (acl) {
215 int error = posix_acl_permission(inode, acl, mask);
216 posix_acl_release(acl);
217 return error;
218 }
84635d68 219#endif
e77819e5
LT		 220
221 return -EAGAIN;
222}
223
5909ccaa 224/*
948409c7 225 * This does the basic permission checking
1da177e4 226 */
7e40145e 227static int acl_permission_check(struct inode *inode, int mask)
1da177e4 228{
26cf46be 229 unsigned int mode = inode->i_mode;
1da177e4 230
e795b717
SH		 231 if (current_user_ns() != inode_userns(inode))
232 goto other_perms;
233
14067ff5 234 if (likely(current_fsuid() == inode->i_uid))
1da177e4
LT		 235 mode >>= 6;
236 else {
e77819e5 237 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
7e40145e 238 int error = check_acl(inode, mask);
b74c79e9
NP		 239 if (error != -EAGAIN)
240 return error;
1da177e4
LT		 241 }
242
243 if (in_group_p(inode->i_gid))
244 mode >>= 3;
245 }
246
e795b717 247other_perms:
1da177e4
LT		 248 /*
249 * If the DACs are ok we don't need any capability check.
250 */
9c2c7039 251 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1da177e4 252 return 0;
5909ccaa
LT		 253 return -EACCES;
254}
255
256/**
b74c79e9 257 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa 258 * @inode: inode to check access rights for
8fd90c8d 259 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
5909ccaa
LT		 260 *
261 * Used to check for read/write/execute permissions on a file.
262 * We use "fsuid" for this, letting us set arbitrary permissions
263 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 264 * are used for other things.
265 *
266 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
267 * request cannot be satisfied (eg. requires blocking or too much complexity).
268 * It would then be called again in ref-walk mode.
5909ccaa 269 */
2830ba7f 270int generic_permission(struct inode *inode, int mask)
5909ccaa
LT		 271{
272 int ret;
273
274 /*
948409c7 275 * Do the basic permission checks.
5909ccaa 276 */
7e40145e 277 ret = acl_permission_check(inode, mask);
5909ccaa
LT		 278 if (ret != -EACCES)
279 return ret;
1da177e4 280
d594e7ec
AV		 281 if (S_ISDIR(inode->i_mode)) {
282 /* DACs are overridable for directories */
283 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
284 return 0;
285 if (!(mask & MAY_WRITE))
286 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
287 return 0;
288 return -EACCES;
289 }
1da177e4
LT		 290 /*
291 * Read/write DACs are always overridable.
d594e7ec
AV		 292 * Executable DACs are overridable when there is
293 * at least one exec bit set.
1da177e4 294 */
d594e7ec 295 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
e795b717 296 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
1da177e4
LT		 297 return 0;
298
299 /*
300 * Searching includes executable on directories, else just read.
301 */
7ea66001 302 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
d594e7ec 303 if (mask == MAY_READ)
e795b717 304 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
1da177e4
LT		 305 return 0;
306
307 return -EACCES;
308}
309
3ddcd056
LT		 310/*
311 * We _really_ want to just do "generic_permission()" without
312 * even looking at the inode->i_op values. So we keep a cache
313 * flag in inode->i_opflags, that says "this has not special
314 * permission function, use the fast case".
315 */
316static inline int do_inode_permission(struct inode *inode, int mask)
317{
318 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
319 if (likely(inode->i_op->permission))
320 return inode->i_op->permission(inode, mask);
321
322 /* This gets set once for the inode lifetime */
323 spin_lock(&inode->i_lock);
324 inode->i_opflags |= IOP_FASTPERM;
325 spin_unlock(&inode->i_lock);
326 }
327 return generic_permission(inode, mask);
328}
329
cb23beb5
CH		 330/**
331 * inode_permission - check for access rights to a given inode
332 * @inode: inode to check permission on
8fd90c8d 333 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
cb23beb5
CH		 334 *
335 * Used to check for read/write/execute permissions on an inode.
336 * We use "fsuid" for this, letting us set arbitrary permissions
337 * for filesystem access without changing the "normal" uids which
338 * are used for other things.
948409c7
AG		 339 *
340 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
cb23beb5 341 */
f419a2e3 342int inode_permission(struct inode *inode, int mask)
1da177e4 343{
e6305c43 344 int retval;
1da177e4 345
3ddcd056 346 if (unlikely(mask & MAY_WRITE)) {
22590e41 347 umode_t mode = inode->i_mode;
1da177e4
LT		 348
349 /*
350 * Nobody gets write access to a read-only fs.
351 */
352 if (IS_RDONLY(inode) &&
353 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
354 return -EROFS;
355
356 /*
357 * Nobody gets write access to an immutable file.
358 */
359 if (IS_IMMUTABLE(inode))
360 return -EACCES;
361 }
362
3ddcd056 363 retval = do_inode_permission(inode, mask);
1da177e4
LT		 364 if (retval)
365 return retval;
366
08ce5f16
SH		 367 retval = devcgroup_inode_permission(inode, mask);
368 if (retval)
369 return retval;
370
d09ca739 371 return security_inode_permission(inode, mask);
1da177e4
LT		 372}
373
5dd784d0
JB		 374/**
375 * path_get - get a reference to a path
376 * @path: path to get the reference to
377 *
378 * Given a path increment the reference count to the dentry and the vfsmount.
379 */
380void path_get(struct path *path)
381{
382 mntget(path->mnt);
383 dget(path->dentry);
384}
385EXPORT_SYMBOL(path_get);
386
1d957f9b
JB		 387/**
388 * path_put - put a reference to a path
389 * @path: path to put the reference to
390 *
391 * Given a path decrement the reference count to the dentry and the vfsmount.
392 */
393void path_put(struct path *path)
1da177e4 394{
1d957f9b
JB		 395 dput(path->dentry);
396 mntput(path->mnt);
1da177e4 397}
1d957f9b 398EXPORT_SYMBOL(path_put);
1da177e4 399
19660af7 400/*
31e6b01f 401 * Path walking has 2 modes, rcu-walk and ref-walk (see
19660af7
AV		 402 * Documentation/filesystems/path-lookup.txt). In situations when we can't
403 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
404 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
405 * mode. Refcounts are grabbed at the last known good point before rcu-walk
406 * got stuck, so ref-walk may continue from there. If this is not successful
407 * (eg. a seqcount has changed), then failure is returned and it's up to caller
408 * to restart the path walk from the beginning in ref-walk mode.
31e6b01f 409 */
31e6b01f
NP		 410
411/**
19660af7
AV		 412 * unlazy_walk - try to switch to ref-walk mode.
413 * @nd: nameidata pathwalk data
414 * @dentry: child of nd->path.dentry or NULL
39191628 415 * Returns: 0 on success, -ECHILD on failure
31e6b01f 416 *
19660af7
AV		 417 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
418 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
419 * @nd or NULL. Must be called from rcu-walk context.
31e6b01f 420 */
19660af7 421static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
31e6b01f
NP		 422{
423 struct fs_struct *fs = current->fs;
424 struct dentry *parent = nd->path.dentry;
5b6ca027 425 int want_root = 0;
31e6b01f
NP		 426
427 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 428 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
429 want_root = 1;
31e6b01f
NP		 430 spin_lock(&fs->lock);
431 if (nd->root.mnt != fs->root.mnt ||
432 nd->root.dentry != fs->root.dentry)
433 goto err_root;
434 }
435 spin_lock(&parent->d_lock);
19660af7
AV		 436 if (!dentry) {
437 if (!__d_rcu_to_refcount(parent, nd->seq))
438 goto err_parent;
439 BUG_ON(nd->inode != parent->d_inode);
440 } else {
94c0d4ec
AV		 441 if (dentry->d_parent != parent)
442 goto err_parent;
19660af7
AV		 443 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
444 if (!__d_rcu_to_refcount(dentry, nd->seq))
445 goto err_child;
446 /*
447 * If the sequence check on the child dentry passed, then
448 * the child has not been removed from its parent. This
449 * means the parent dentry must be valid and able to take
450 * a reference at this point.
451 */
452 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
453 BUG_ON(!parent->d_count);
454 parent->d_count++;
455 spin_unlock(&dentry->d_lock);
456 }
31e6b01f 457 spin_unlock(&parent->d_lock);
5b6ca027 458 if (want_root) {
31e6b01f
NP		 459 path_get(&nd->root);
460 spin_unlock(&fs->lock);
461 }
462 mntget(nd->path.mnt);
463
464 rcu_read_unlock();
465 br_read_unlock(vfsmount_lock);
466 nd->flags &= ~LOOKUP_RCU;
467 return 0;
19660af7
AV		 468
469err_child:
31e6b01f 470 spin_unlock(&dentry->d_lock);
19660af7 471err_parent:
31e6b01f
NP		 472 spin_unlock(&parent->d_lock);
473err_root:
5b6ca027 474 if (want_root)
31e6b01f
NP		 475 spin_unlock(&fs->lock);
476 return -ECHILD;
477}
478
834f2a4a
TM		 479/**
480 * release_open_intent - free up open intent resources
481 * @nd: pointer to nameidata
482 */
483void release_open_intent(struct nameidata *nd)
484{
2dab5974
LT		 485 struct file *file = nd->intent.open.file;
486
487 if (file && !IS_ERR(file)) {
488 if (file->f_path.dentry == NULL)
489 put_filp(file);
490 else
491 fput(file);
492 }
834f2a4a
TM		 493}
494
f60aef7e 495static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 496{
f60aef7e 497 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 498}
499
9f1fafee
AV		 500/**
501 * complete_walk - successful completion of path walk
502 * @nd: pointer nameidata
39159de2 503 *
9f1fafee
AV		 504 * If we had been in RCU mode, drop out of it and legitimize nd->path.
505 * Revalidate the final result, unless we'd already done that during
506 * the path walk or the filesystem doesn't ask for it. Return 0 on
507 * success, -error on failure. In case of failure caller does not
508 * need to drop nd->path.
39159de2 509 */
9f1fafee 510static int complete_walk(struct nameidata *nd)
39159de2 511{
16c2cd71 512 struct dentry *dentry = nd->path.dentry;
39159de2 513 int status;
39159de2 514
9f1fafee
AV		 515 if (nd->flags & LOOKUP_RCU) {
516 nd->flags &= ~LOOKUP_RCU;
517 if (!(nd->flags & LOOKUP_ROOT))
518 nd->root.mnt = NULL;
519 spin_lock(&dentry->d_lock);
520 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
521 spin_unlock(&dentry->d_lock);
522 rcu_read_unlock();
523 br_read_unlock(vfsmount_lock);
524 return -ECHILD;
525 }
526 BUG_ON(nd->inode != dentry->d_inode);
527 spin_unlock(&dentry->d_lock);
528 mntget(nd->path.mnt);
529 rcu_read_unlock();
530 br_read_unlock(vfsmount_lock);
531 }
532
16c2cd71
AV		 533 if (likely(!(nd->flags & LOOKUP_JUMPED)))
534 return 0;
535
536 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 537 return 0;
538
16c2cd71
AV		 539 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
540 return 0;
541
542 /* Note: we do not d_invalidate() */
34286d66 543 status = d_revalidate(dentry, nd);
39159de2
JL		 544 if (status > 0)
545 return 0;
546
16c2cd71 547 if (!status)
39159de2 548 status = -ESTALE;
16c2cd71 549
9f1fafee 550 path_put(&nd->path);
39159de2
JL		 551 return status;
552}
553
2a737871
AV		 554static __always_inline void set_root(struct nameidata *nd)
555{
f7ad3c6b
MS		 556 if (!nd->root.mnt)
557 get_fs_root(current->fs, &nd->root);
2a737871
AV		 558}
559
6de88d72
AV		 560static int link_path_walk(const char *, struct nameidata *);
561
31e6b01f
NP		 562static __always_inline void set_root_rcu(struct nameidata *nd)
563{
564 if (!nd->root.mnt) {
565 struct fs_struct *fs = current->fs;
c28cc364
NP		 566 unsigned seq;
567
568 do {
569 seq = read_seqcount_begin(&fs->seq);
570 nd->root = fs->root;
c1530019 571 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
c28cc364 572 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 573 }
574}
575
f1662356 576static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 577{
31e6b01f
NP		 578 int ret;
579
1da177e4
LT		 580 if (IS_ERR(link))
581 goto fail;
582
583 if (*link == '/') {
2a737871 584 set_root(nd);
1d957f9b 585 path_put(&nd->path);
2a737871
AV		 586 nd->path = nd->root;
587 path_get(&nd->root);
16c2cd71 588 nd->flags |= LOOKUP_JUMPED;
1da177e4 589 }
31e6b01f 590 nd->inode = nd->path.dentry->d_inode;
b4091d5f 591
31e6b01f
NP		 592 ret = link_path_walk(link, nd);
593 return ret;
1da177e4 594fail:
1d957f9b 595 path_put(&nd->path);
1da177e4
LT		 596 return PTR_ERR(link);
597}
598
1d957f9b 599static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 600{
601 dput(path->dentry);
4ac91378 602 if (path->mnt != nd->path.mnt)
051d3812
IK		 603 mntput(path->mnt);
604}
605
7b9337aa
NP		 606static inline void path_to_nameidata(const struct path *path,
607 struct nameidata *nd)
051d3812 608{
31e6b01f
NP		 609 if (!(nd->flags & LOOKUP_RCU)) {
610 dput(nd->path.dentry);
611 if (nd->path.mnt != path->mnt)
612 mntput(nd->path.mnt);
9a229683 613 }
31e6b01f 614 nd->path.mnt = path->mnt;
4ac91378 615 nd->path.dentry = path->dentry;
051d3812
IK		 616}
617
574197e0
AV		 618static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
619{
620 struct inode *inode = link->dentry->d_inode;
621 if (!IS_ERR(cookie) && inode->i_op->put_link)
622 inode->i_op->put_link(link->dentry, nd, cookie);
623 path_put(link);
624}
625
def4af30 626static __always_inline int
574197e0 627follow_link(struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 628{
629 int error;
7b9337aa 630 struct dentry *dentry = link->dentry;
1da177e4 631
844a3917
AV		 632 BUG_ON(nd->flags & LOOKUP_RCU);
633
0e794589
AV		 634 if (link->mnt == nd->path.mnt)
635 mntget(link->mnt);
636
574197e0
AV		 637 if (unlikely(current->total_link_count >= 40)) {
638 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
574197e0
AV		 639 path_put(&nd->path);
640 return -ELOOP;
641 }
642 cond_resched();
643 current->total_link_count++;
644
68ac1234 645 touch_atime(link);
1da177e4 646 nd_set_link(nd, NULL);
cd4e91d3 647
36f3b4f6
AV		 648 error = security_inode_follow_link(link->dentry, nd);
649 if (error) {
650 *p = ERR_PTR(error); /* no ->put_link(), please */
651 path_put(&nd->path);
652 return error;
653 }
654
86acdca1 655 nd->last_type = LAST_BIND;
def4af30
AV		 656 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
657 error = PTR_ERR(*p);
658 if (!IS_ERR(*p)) {
1da177e4 659 char *s = nd_get_link(nd);
cc314eef 660 error = 0;
1da177e4
LT		 661 if (s)
662 error = __vfs_follow_link(nd, s);
bcda7652 663 else if (nd->last_type == LAST_BIND) {
16c2cd71 664 nd->flags |= LOOKUP_JUMPED;
b21041d0
AV		 665 nd->inode = nd->path.dentry->d_inode;
666 if (nd->inode->i_op->follow_link) {
bcda7652
AV		 667 /* stepped on a _really_ weird one */
668 path_put(&nd->path);
669 error = -ELOOP;
670 }
671 }
1da177e4 672 }
1da177e4
LT		 673 return error;
674}
675
31e6b01f
NP		 676static int follow_up_rcu(struct path *path)
677{
0714a533
AV		 678 struct mount *mnt = real_mount(path->mnt);
679 struct mount *parent;
31e6b01f
NP		 680 struct dentry *mountpoint;
681
0714a533
AV		 682 parent = mnt->mnt_parent;
683 if (&parent->mnt == path->mnt)
31e6b01f 684 return 0;
a73324da 685 mountpoint = mnt->mnt_mountpoint;
31e6b01f 686 path->dentry = mountpoint;
0714a533 687 path->mnt = &parent->mnt;
31e6b01f
NP		 688 return 1;
689}
690
bab77ebf 691int follow_up(struct path *path)
1da177e4 692{
0714a533
AV		 693 struct mount *mnt = real_mount(path->mnt);
694 struct mount *parent;
1da177e4 695 struct dentry *mountpoint;
99b7db7b
NP		 696
697 br_read_lock(vfsmount_lock);
0714a533
AV		 698 parent = mnt->mnt_parent;
699 if (&parent->mnt == path->mnt) {
99b7db7b 700 br_read_unlock(vfsmount_lock);
1da177e4
LT		 701 return 0;
702 }
0714a533 703 mntget(&parent->mnt);
a73324da 704 mountpoint = dget(mnt->mnt_mountpoint);
99b7db7b 705 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 706 dput(path->dentry);
707 path->dentry = mountpoint;
708 mntput(path->mnt);
0714a533 709 path->mnt = &parent->mnt;
1da177e4
LT		 710 return 1;
711}
712
b5c84bf6 713/*
9875cf80
DH		 714 * Perform an automount
715 * - return -EISDIR to tell follow_managed() to stop and return the path we
716 * were called with.
1da177e4 717 */
9875cf80
DH		 718static int follow_automount(struct path *path, unsigned flags,
719 bool *need_mntput)
31e6b01f 720{
9875cf80 721 struct vfsmount *mnt;
ea5b778a 722 int err;
9875cf80
DH		 723
724 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
725 return -EREMOTE;
726
0ec26fd0
MS		 727 /* We don't want to mount if someone's just doing a stat -
728 * unless they're stat'ing a directory and appended a '/' to
729 * the name.
730 *
731 * We do, however, want to mount if someone wants to open or
732 * create a file of any type under the mountpoint, wants to
733 * traverse through the mountpoint or wants to open the
734 * mounted directory. Also, autofs may mark negative dentries
735 * as being automount points. These will need the attentions
736 * of the daemon to instantiate them before they can be used.
9875cf80 737 */
0ec26fd0 738 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
d94c177b 739 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
0ec26fd0
MS		 740 path->dentry->d_inode)
741 return -EISDIR;
742
9875cf80
DH		 743 current->total_link_count++;
744 if (current->total_link_count >= 40)
745 return -ELOOP;
746
747 mnt = path->dentry->d_op->d_automount(path);
748 if (IS_ERR(mnt)) {
749 /*
750 * The filesystem is allowed to return -EISDIR here to indicate
751 * it doesn't want to automount. For instance, autofs would do
752 * this so that its userspace daemon can mount on this dentry.
753 *
754 * However, we can only permit this if it's a terminal point in
755 * the path being looked up; if it wasn't then the remainder of
756 * the path is inaccessible and we should say so.
757 */
49084c3b 758 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
9875cf80
DH		 759 return -EREMOTE;
760 return PTR_ERR(mnt);
31e6b01f 761 }
ea5b778a 762
9875cf80
DH		 763 if (!mnt) /* mount collision */
764 return 0;
31e6b01f 765
8aef1884
AV		 766 if (!*need_mntput) {
767 /* lock_mount() may release path->mnt on error */
768 mntget(path->mnt);
769 *need_mntput = true;
770 }
19a167af 771 err = finish_automount(mnt, path);
9875cf80 772
ea5b778a
DH		 773 switch (err) {
774 case -EBUSY:
775 /* Someone else made a mount here whilst we were busy */
19a167af 776 return 0;
ea5b778a 777 case 0:
8aef1884 778 path_put(path);
ea5b778a
DH		 779 path->mnt = mnt;
780 path->dentry = dget(mnt->mnt_root);
ea5b778a 781 return 0;
19a167af
AV		 782 default:
783 return err;
ea5b778a 784 }
19a167af 785
463ffb2e
AV		 786}
787
9875cf80
DH		 788/*
789 * Handle a dentry that is managed in some way.
cc53ce53 790 * - Flagged for transit management (autofs)
9875cf80
DH		 791 * - Flagged as mountpoint
792 * - Flagged as automount point
793 *
794 * This may only be called in refwalk mode.
795 *
796 * Serialization is taken care of in namespace.c
797 */
798static int follow_managed(struct path *path, unsigned flags)
1da177e4 799{
8aef1884 800 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
9875cf80
DH		 801 unsigned managed;
802 bool need_mntput = false;
8aef1884 803 int ret = 0;
9875cf80
DH		 804
805 /* Given that we're not holding a lock here, we retain the value in a
806 * local variable for each dentry as we look at it so that we don't see
807 * the components of that value change under us */
808 while (managed = ACCESS_ONCE(path->dentry->d_flags),
809 managed &= DCACHE_MANAGED_DENTRY,
810 unlikely(managed != 0)) {
cc53ce53
DH		 811 /* Allow the filesystem to manage the transit without i_mutex
812 * being held. */
813 if (managed & DCACHE_MANAGE_TRANSIT) {
814 BUG_ON(!path->dentry->d_op);
815 BUG_ON(!path->dentry->d_op->d_manage);
1aed3e42 816 ret = path->dentry->d_op->d_manage(path->dentry, false);
cc53ce53 817 if (ret < 0)
8aef1884 818 break;
cc53ce53
DH		 819 }
820
9875cf80
DH		 821 /* Transit to a mounted filesystem. */
822 if (managed & DCACHE_MOUNTED) {
823 struct vfsmount *mounted = lookup_mnt(path);
824 if (mounted) {
825 dput(path->dentry);
826 if (need_mntput)
827 mntput(path->mnt);
828 path->mnt = mounted;
829 path->dentry = dget(mounted->mnt_root);
830 need_mntput = true;
831 continue;
832 }
833
834 /* Something is mounted on this dentry in another
835 * namespace and/or whatever was mounted there in this
836 * namespace got unmounted before we managed to get the
837 * vfsmount_lock */
838 }
839
840 /* Handle an automount point */
841 if (managed & DCACHE_NEED_AUTOMOUNT) {
842 ret = follow_automount(path, flags, &need_mntput);
843 if (ret < 0)
8aef1884 844 break;
9875cf80
DH		 845 continue;
846 }
847
848 /* We didn't change the current path point */
849 break;
1da177e4 850 }
8aef1884
AV		 851
852 if (need_mntput && path->mnt == mnt)
853 mntput(path->mnt);
854 if (ret == -EISDIR)
855 ret = 0;
a3fbbde7 856 return ret < 0 ? ret : need_mntput;
1da177e4
LT		 857}
858
cc53ce53 859int follow_down_one(struct path *path)
1da177e4
LT		 860{
861 struct vfsmount *mounted;
862
1c755af4 863 mounted = lookup_mnt(path);
1da177e4 864 if (mounted) {
9393bd07
AV		 865 dput(path->dentry);
866 mntput(path->mnt);
867 path->mnt = mounted;
868 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 869 return 1;
870 }
871 return 0;
872}
873
62a7375e
IK		 874static inline bool managed_dentry_might_block(struct dentry *dentry)
875{
876 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
877 dentry->d_op->d_manage(dentry, true) < 0);
878}
879
9875cf80 880/*
287548e4
AV		 881 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
882 * we meet a managed dentry that would need blocking.
9875cf80
DH		 883 */
884static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
287548e4 885 struct inode **inode)
9875cf80 886{
62a7375e 887 for (;;) {
c7105365 888 struct mount *mounted;
62a7375e
IK		 889 /*
890 * Don't forget we might have a non-mountpoint managed dentry
891 * that wants to block transit.
892 */
287548e4 893 if (unlikely(managed_dentry_might_block(path->dentry)))
ab90911f 894 return false;
62a7375e
IK		 895
896 if (!d_mountpoint(path->dentry))
897 break;
898
9875cf80
DH		 899 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
900 if (!mounted)
901 break;
c7105365
AV		 902 path->mnt = &mounted->mnt;
903 path->dentry = mounted->mnt.mnt_root;
a3fbbde7 904 nd->flags |= LOOKUP_JUMPED;
9875cf80 905 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
59430262
LT		 906 /*
907 * Update the inode too. We don't need to re-check the
908 * dentry sequence number here after this d_inode read,
909 * because a mount-point is always pinned.
910 */
911 *inode = path->dentry->d_inode;
9875cf80 912 }
9875cf80
DH		 913 return true;
914}
915
dea39376 916static void follow_mount_rcu(struct nameidata *nd)
287548e4 917{
dea39376 918 while (d_mountpoint(nd->path.dentry)) {
c7105365 919 struct mount *mounted;
dea39376 920 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
287548e4
AV		 921 if (!mounted)
922 break;
c7105365
AV		 923 nd->path.mnt = &mounted->mnt;
924 nd->path.dentry = mounted->mnt.mnt_root;
dea39376 925 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
287548e4
AV		 926 }
927}
928
31e6b01f
NP		 929static int follow_dotdot_rcu(struct nameidata *nd)
930{
31e6b01f
NP		 931 set_root_rcu(nd);
932
9875cf80 933 while (1) {
31e6b01f
NP		 934 if (nd->path.dentry == nd->root.dentry &&
935 nd->path.mnt == nd->root.mnt) {
936 break;
937 }
938 if (nd->path.dentry != nd->path.mnt->mnt_root) {
939 struct dentry *old = nd->path.dentry;
940 struct dentry *parent = old->d_parent;
941 unsigned seq;
942
943 seq = read_seqcount_begin(&parent->d_seq);
944 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 945 goto failed;
31e6b01f
NP		 946 nd->path.dentry = parent;
947 nd->seq = seq;
948 break;
949 }
950 if (!follow_up_rcu(&nd->path))
951 break;
952 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f 953 }
dea39376
AV		 954 follow_mount_rcu(nd);
955 nd->inode = nd->path.dentry->d_inode;
31e6b01f 956 return 0;
ef7562d5
AV		 957
958failed:
959 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 960 if (!(nd->flags & LOOKUP_ROOT))
961 nd->root.mnt = NULL;
ef7562d5
AV		 962 rcu_read_unlock();
963 br_read_unlock(vfsmount_lock);
964 return -ECHILD;
31e6b01f
NP		 965}
966
cc53ce53
DH		 967/*
968 * Follow down to the covering mount currently visible to userspace. At each
969 * point, the filesystem owning that dentry may be queried as to whether the
970 * caller is permitted to proceed or not.
cc53ce53 971 */
7cc90cc3 972int follow_down(struct path *path)
cc53ce53
DH		 973{
974 unsigned managed;
975 int ret;
976
977 while (managed = ACCESS_ONCE(path->dentry->d_flags),
978 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
979 /* Allow the filesystem to manage the transit without i_mutex
980 * being held.
981 *
982 * We indicate to the filesystem if someone is trying to mount
983 * something here. This gives autofs the chance to deny anyone
984 * other than its daemon the right to mount on its
985 * superstructure.
986 *
987 * The filesystem may sleep at this point.
988 */
989 if (managed & DCACHE_MANAGE_TRANSIT) {
990 BUG_ON(!path->dentry->d_op);
991 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 992 ret = path->dentry->d_op->d_manage(
1aed3e42 993 path->dentry, false);
cc53ce53
DH		 994 if (ret < 0)
995 return ret == -EISDIR ? 0 : ret;
996 }
997
998 /* Transit to a mounted filesystem. */
999 if (managed & DCACHE_MOUNTED) {
1000 struct vfsmount *mounted = lookup_mnt(path);
1001 if (!mounted)
1002 break;
1003 dput(path->dentry);
1004 mntput(path->mnt);
1005 path->mnt = mounted;
1006 path->dentry = dget(mounted->mnt_root);
1007 continue;
1008 }
1009
1010 /* Don't handle automount points here */
1011 break;
1012 }
1013 return 0;
1014}
1015
9875cf80
DH		 1016/*
1017 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1018 */
1019static void follow_mount(struct path *path)
1020{
1021 while (d_mountpoint(path->dentry)) {
1022 struct vfsmount *mounted = lookup_mnt(path);
1023 if (!mounted)
1024 break;
1025 dput(path->dentry);
1026 mntput(path->mnt);
1027 path->mnt = mounted;
1028 path->dentry = dget(mounted->mnt_root);
1029 }
1030}
1031
31e6b01f 1032static void follow_dotdot(struct nameidata *nd)
1da177e4 1033{
2a737871 1034 set_root(nd);
e518ddb7 1035
1da177e4 1036 while(1) {
4ac91378 1037 struct dentry *old = nd->path.dentry;
1da177e4 1038
2a737871
AV		 1039 if (nd->path.dentry == nd->root.dentry &&
1040 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1041 break;
1042 }
4ac91378 1043 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1044 /* rare case of legitimate dget_parent()... */
1045 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1046 dput(old);
1047 break;
1048 }
3088dd70 1049 if (!follow_up(&nd->path))
1da177e4 1050 break;
1da177e4 1051 }
79ed0226 1052 follow_mount(&nd->path);
31e6b01f 1053 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1054}
1055
baa03890
NP		 1056/*
1057 * Allocate a dentry with name and parent, and perform a parent
1058 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1059 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1060 * have verified that no child exists while under i_mutex.
1061 */
1062static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1063 struct qstr *name, struct nameidata *nd)
1064{
1065 struct inode *inode = parent->d_inode;
1066 struct dentry *dentry;
1067 struct dentry *old;
1068
1069 /* Don't create child dentry for a dead directory. */
1070 if (unlikely(IS_DEADDIR(inode)))
1071 return ERR_PTR(-ENOENT);
1072
1073 dentry = d_alloc(parent, name);
1074 if (unlikely(!dentry))
1075 return ERR_PTR(-ENOMEM);
1076
1077 old = inode->i_op->lookup(inode, dentry, nd);
1078 if (unlikely(old)) {
1079 dput(dentry);
1080 dentry = old;
1081 }
1082 return dentry;
1083}
1084
44396f4b
JB		 1085/*
1086 * We already have a dentry, but require a lookup to be performed on the parent
1087 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1088 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1089 * child exists while under i_mutex.
1090 */
1091static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1092 struct nameidata *nd)
1093{
1094 struct inode *inode = parent->d_inode;
1095 struct dentry *old;
1096
1097 /* Don't create child dentry for a dead directory. */
e188dc02
MS		 1098 if (unlikely(IS_DEADDIR(inode))) {
1099 dput(dentry);
44396f4b 1100 return ERR_PTR(-ENOENT);
e188dc02 1101 }
44396f4b
JB		 1102
1103 old = inode->i_op->lookup(inode, dentry, nd);
1104 if (unlikely(old)) {
1105 dput(dentry);
1106 dentry = old;
1107 }
1108 return dentry;
1109}
1110
a3255546
AV		 1111static struct dentry *__lookup_hash(struct qstr *name,
1112 struct dentry *base, struct nameidata *nd)
1113{
1114 struct dentry *dentry;
1115
1116 /*
1117 * Don't bother with __d_lookup: callers are for creat as
1118 * well as unlink, so a lot of the time it would cost
1119 * a double lookup.
1120 */
1121 dentry = d_lookup(base, name);
1122
1123 if (dentry && d_need_lookup(dentry)) {
1124 /*
1125 * __lookup_hash is called with the parent dir's i_mutex already
1126 * held, so we are good to go here.
1127 */
1128 return d_inode_lookup(base, dentry, nd);
1129 }
1130
1131 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1132 int status = d_revalidate(dentry, nd);
1133 if (unlikely(status <= 0)) {
1134 /*
1135 * The dentry failed validation.
1136 * If d_revalidate returned 0 attempt to invalidate
1137 * the dentry otherwise d_revalidate is asking us
1138 * to return a fail status.
1139 */
1140 if (status < 0) {
1141 dput(dentry);
1142 return ERR_PTR(status);
1143 } else if (!d_invalidate(dentry)) {
1144 dput(dentry);
1145 dentry = NULL;
1146 }
1147 }
1148 }
1149
1150 if (!dentry)
1151 dentry = d_alloc_and_lookup(base, name, nd);
1152
1153 return dentry;
1154}
1155
1da177e4
LT		 1156/*
1157 * It's more convoluted than I'd like it to be, but... it's still fairly
1158 * small and for now I'd prefer to have fast path as straight as possible.
1159 * It _is_ time-critical.
1160 */
1161static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1162 struct path *path, struct inode **inode)
1da177e4 1163{
4ac91378 1164 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1165 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1166 int need_reval = 1;
1167 int status = 1;
9875cf80
DH		 1168 int err;
1169
b04f784e
NP		 1170 /*
1171 * Rename seqlock is not required here because in the off chance
1172 * of a false negative due to a concurrent rename, we're going to
1173 * do the non-racy lookup, below.
1174 */
31e6b01f
NP		 1175 if (nd->flags & LOOKUP_RCU) {
1176 unsigned seq;
31e6b01f
NP		 1177 *inode = nd->inode;
1178 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1179 if (!dentry)
1180 goto unlazy;
1181
31e6b01f
NP		 1182 /* Memory barrier in read_seqcount_begin of child is enough */
1183 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1184 return -ECHILD;
31e6b01f 1185 nd->seq = seq;
5a18fff2 1186
fa4ee159
MS		 1187 if (unlikely(d_need_lookup(dentry)))
1188 goto unlazy;
24643087 1189 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1190 status = d_revalidate(dentry, nd);
1191 if (unlikely(status <= 0)) {
1192 if (status != -ECHILD)
1193 need_reval = 0;
1194 goto unlazy;
1195 }
24643087 1196 }
31e6b01f
NP		 1197 path->mnt = mnt;
1198 path->dentry = dentry;
d6e9bd25
AV		 1199 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1200 goto unlazy;
1201 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1202 goto unlazy;
1203 return 0;
5a18fff2 1204unlazy:
19660af7
AV		 1205 if (unlazy_walk(nd, dentry))
1206 return -ECHILD;
5a18fff2
AV		 1207 } else {
1208 dentry = __d_lookup(parent, name);
9875cf80 1209 }
5a18fff2 1210
44396f4b
JB		 1211 if (dentry && unlikely(d_need_lookup(dentry))) {
1212 dput(dentry);
1213 dentry = NULL;
1214 }
5a18fff2
AV		 1215retry:
1216 if (unlikely(!dentry)) {
1217 struct inode *dir = parent->d_inode;
1218 BUG_ON(nd->inode != dir);
1219
1220 mutex_lock(&dir->i_mutex);
a3255546 1221 dentry = __lookup_hash(name, parent, nd);
37c17e1f 1222 mutex_unlock(&dir->i_mutex);
ec335e91
AV		 1223 if (IS_ERR(dentry))
1224 return PTR_ERR(dentry);
3f6c7c71 1225 goto done;
5a18fff2
AV		 1226 }
1227 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1228 status = d_revalidate(dentry, nd);
1229 if (unlikely(status <= 0)) {
1230 if (status < 0) {
1231 dput(dentry);
1232 return status;
1233 }
1234 if (!d_invalidate(dentry)) {
1235 dput(dentry);
1236 dentry = NULL;
5a18fff2
AV		 1237 goto retry;
1238 }
24643087 1239 }
3f6c7c71 1240done:
9875cf80
DH		 1241 path->mnt = mnt;
1242 path->dentry = dentry;
1243 err = follow_managed(path, nd->flags);
89312214
IK		 1244 if (unlikely(err < 0)) {
1245 path_put_conditional(path, nd);
9875cf80 1246 return err;
89312214 1247 }
a3fbbde7
AV		 1248 if (err)
1249 nd->flags |= LOOKUP_JUMPED;
9875cf80 1250 *inode = path->dentry->d_inode;
1da177e4 1251 return 0;
1da177e4
LT		 1252}
1253
52094c8a
AV		 1254static inline int may_lookup(struct nameidata *nd)
1255{
1256 if (nd->flags & LOOKUP_RCU) {
4ad5abb3 1257 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
52094c8a
AV		 1258 if (err != -ECHILD)
1259 return err;
19660af7 1260 if (unlazy_walk(nd, NULL))
52094c8a
AV		 1261 return -ECHILD;
1262 }
4ad5abb3 1263 return inode_permission(nd->inode, MAY_EXEC);
52094c8a
AV		 1264}
1265
9856fa1b
AV		 1266static inline int handle_dots(struct nameidata *nd, int type)
1267{
1268 if (type == LAST_DOTDOT) {
1269 if (nd->flags & LOOKUP_RCU) {
1270 if (follow_dotdot_rcu(nd))
1271 return -ECHILD;
1272 } else
1273 follow_dotdot(nd);
1274 }
1275 return 0;
1276}
1277
951361f9
AV		 1278static void terminate_walk(struct nameidata *nd)
1279{
1280 if (!(nd->flags & LOOKUP_RCU)) {
1281 path_put(&nd->path);
1282 } else {
1283 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1284 if (!(nd->flags & LOOKUP_ROOT))
1285 nd->root.mnt = NULL;
951361f9
AV		 1286 rcu_read_unlock();
1287 br_read_unlock(vfsmount_lock);
1288 }
1289}
1290
3ddcd056
LT		 1291/*
1292 * Do we need to follow links? We _really_ want to be able
1293 * to do this check without having to look at inode->i_op,
1294 * so we keep a cache of "no, this doesn't need follow_link"
1295 * for the common case.
1296 */
7813b94a 1297static inline int should_follow_link(struct inode *inode, int follow)
3ddcd056
LT		 1298{
1299 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1300 if (likely(inode->i_op->follow_link))
1301 return follow;
1302
1303 /* This gets set once for the inode lifetime */
1304 spin_lock(&inode->i_lock);
1305 inode->i_opflags |= IOP_NOFOLLOW;
1306 spin_unlock(&inode->i_lock);
1307 }
1308 return 0;
1309}
1310
ce57dfc1
AV		 1311static inline int walk_component(struct nameidata *nd, struct path *path,
1312 struct qstr *name, int type, int follow)
1313{
1314 struct inode *inode;
1315 int err;
1316 /*
1317 * "." and ".." are special - ".." especially so because it has
1318 * to be able to know about the current root directory and
1319 * parent relationships.
1320 */
1321 if (unlikely(type != LAST_NORM))
1322 return handle_dots(nd, type);
1323 err = do_lookup(nd, name, path, &inode);
1324 if (unlikely(err)) {
1325 terminate_walk(nd);
1326 return err;
1327 }
1328 if (!inode) {
1329 path_to_nameidata(path, nd);
1330 terminate_walk(nd);
1331 return -ENOENT;
1332 }
7813b94a 1333 if (should_follow_link(inode, follow)) {
19660af7
AV		 1334 if (nd->flags & LOOKUP_RCU) {
1335 if (unlikely(unlazy_walk(nd, path->dentry))) {
1336 terminate_walk(nd);
1337 return -ECHILD;
1338 }
1339 }
ce57dfc1
AV		 1340 BUG_ON(inode != path->dentry->d_inode);
1341 return 1;
1342 }
1343 path_to_nameidata(path, nd);
1344 nd->inode = inode;
1345 return 0;
1346}
1347
b356379a
AV		 1348/*
1349 * This limits recursive symlink follows to 8, while
1350 * limiting consecutive symlinks to 40.
1351 *
1352 * Without that kind of total limit, nasty chains of consecutive
1353 * symlinks can cause almost arbitrarily long lookups.
1354 */
1355static inline int nested_symlink(struct path *path, struct nameidata *nd)
1356{
1357 int res;
1358
b356379a
AV		 1359 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1360 path_put_conditional(path, nd);
1361 path_put(&nd->path);
1362 return -ELOOP;
1363 }
1a4022f8 1364 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
b356379a
AV		 1365
1366 nd->depth++;
1367 current->link_count++;
1368
1369 do {
1370 struct path link = *path;
1371 void *cookie;
574197e0
AV		 1372
1373 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1374 if (!res)
1375 res = walk_component(nd, path, &nd->last,
1376 nd->last_type, LOOKUP_FOLLOW);
574197e0 1377 put_link(nd, &link, cookie);
b356379a
AV		 1378 } while (res > 0);
1379
1380 current->link_count--;
1381 nd->depth--;
1382 return res;
1383}
1384
3ddcd056
LT		 1385/*
1386 * We really don't want to look at inode->i_op->lookup
1387 * when we don't have to. So we keep a cache bit in
1388 * the inode ->i_opflags field that says "yes, we can
1389 * do lookup on this inode".
1390 */
1391static inline int can_lookup(struct inode *inode)
1392{
1393 if (likely(inode->i_opflags & IOP_LOOKUP))
1394 return 1;
1395 if (likely(!inode->i_op->lookup))
1396 return 0;
1397
1398 /* We do this once for the lifetime of the inode */
1399 spin_lock(&inode->i_lock);
1400 inode->i_opflags |= IOP_LOOKUP;
1401 spin_unlock(&inode->i_lock);
1402 return 1;
1403}
1404
bfcfaa77
LT		 1405/*
1406 * We can do the critical dentry name comparison and hashing
1407 * operations one word at a time, but we are limited to:
1408 *
1409 * - Architectures with fast unaligned word accesses. We could
1410 * do a "get_unaligned()" if this helps and is sufficiently
1411 * fast.
1412 *
1413 * - Little-endian machines (so that we can generate the mask
1414 * of low bytes efficiently). Again, we *could* do a byte
1415 * swapping load on big-endian architectures if that is not
1416 * expensive enough to make the optimization worthless.
1417 *
1418 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1419 * do not trap on the (extremely unlikely) case of a page
1420 * crossing operation.
1421 *
1422 * - Furthermore, we need an efficient 64-bit compile for the
1423 * 64-bit case in order to generate the "number of bytes in
1424 * the final mask". Again, that could be replaced with a
1425 * efficient population count instruction or similar.
1426 */
1427#ifdef CONFIG_DCACHE_WORD_ACCESS
1428
1429#ifdef CONFIG_64BIT
1430
1431/*
1432 * Jan Achrenius on G+: microoptimized version of
1433 * the simpler "(mask & ONEBYTES) * ONEBYTES >> 56"
1434 * that works for the bytemasks without having to
1435 * mask them first.
1436 */
1437static inline long count_masked_bytes(unsigned long mask)
1438{
f7493e5d 1439 return mask*0x0001020304050608ul >> 56;
bfcfaa77
LT		 1440}
1441
1442static inline unsigned int fold_hash(unsigned long hash)
1443{
1444 hash += hash >> (8*sizeof(int));
1445 return hash;
1446}
1447
1448#else /* 32-bit case */
1449
1450/* Carl Chatfield / Jan Achrenius G+ version for 32-bit */
1451static inline long count_masked_bytes(long mask)
1452{
1453 /* (000000 0000ff 00ffff ffffff) -> ( 1 1 2 3 ) */
1454 long a = (0x0ff0001+mask) >> 23;
1455 /* Fix the 1 for 00 case */
1456 return a & mask;
1457}
1458
1459#define fold_hash(x) (x)
1460
1461#endif
1462
1463unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1464{
1465 unsigned long a, mask;
1466 unsigned long hash = 0;
1467
1468 for (;;) {
1469 a = *(unsigned long *)name;
bfcfaa77
LT		 1470 if (len < sizeof(unsigned long))
1471 break;
1472 hash += a;
f132c5be 1473 hash *= 9;
bfcfaa77
LT		 1474 name += sizeof(unsigned long);
1475 len -= sizeof(unsigned long);
1476 if (!len)
1477 goto done;
1478 }
1479 mask = ~(~0ul << len*8);
1480 hash += mask & a;
1481done:
1482 return fold_hash(hash);
1483}
1484EXPORT_SYMBOL(full_name_hash);
1485
989412bb
LT		 1486#define REPEAT_BYTE(x) ((~0ul / 0xff) * (x))
1487#define ONEBYTES REPEAT_BYTE(0x01)
1488#define SLASHBYTES REPEAT_BYTE('/')
1489#define HIGHBITS REPEAT_BYTE(0x80)
bfcfaa77
LT		 1490
1491/* Return the high bit set in the first byte that is a zero */
1492static inline unsigned long has_zero(unsigned long a)
1493{
1494 return ((a - ONEBYTES) & ~a) & HIGHBITS;
1495}
1496
1497/*
1498 * Calculate the length and hash of the path component, and
1499 * return the length of the component;
1500 */
1501static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1502{
1503 unsigned long a, mask, hash, len;
1504
1505 hash = a = 0;
1506 len = -sizeof(unsigned long);
1507 do {
1508 hash = (hash + a) * 9;
1509 len += sizeof(unsigned long);
1510 a = *(unsigned long *)(name+len);
1511 /* Do we have any NUL or '/' bytes in this word? */
1512 mask = has_zero(a) | has_zero(a ^ SLASHBYTES);
1513 } while (!mask);
1514
1515 /* The mask *below* the first high bit set */
1516 mask = (mask - 1) & ~mask;
1517 mask >>= 7;
1518 hash += a & mask;
1519 *hashp = fold_hash(hash);
1520
1521 return len + count_masked_bytes(mask);
1522}
1523
1524#else
1525
0145acc2
LT		 1526unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1527{
1528 unsigned long hash = init_name_hash();
1529 while (len--)
1530 hash = partial_name_hash(*name++, hash);
1531 return end_name_hash(hash);
1532}
ae942ae7 1533EXPORT_SYMBOL(full_name_hash);
0145acc2 1534
200e9ef7
LT		 1535/*
1536 * We know there's a real path component here of at least
1537 * one character.
1538 */
1539static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1540{
1541 unsigned long hash = init_name_hash();
1542 unsigned long len = 0, c;
1543
1544 c = (unsigned char)*name;
1545 do {
1546 len++;
1547 hash = partial_name_hash(c, hash);
1548 c = (unsigned char)name[len];
1549 } while (c && c != '/');
1550 *hashp = end_name_hash(hash);
1551 return len;
1552}
1553
bfcfaa77
LT		 1554#endif
1555
1da177e4
LT		 1556/*
1557 * Name resolution.
ea3834d9
PM		 1558 * This is the basic name resolution function, turning a pathname into
1559 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1560 *
ea3834d9
PM		 1561 * Returns 0 and nd will have valid dentry and mnt on success.
1562 * Returns error and drops reference to input namei data on failure.
1da177e4 1563 */
6de88d72 1564static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1565{
1566 struct path next;
1da177e4 1567 int err;
1da177e4
LT		 1568
1569 while (*name=='/')
1570 name++;
1571 if (!*name)
086e183a 1572 return 0;
1da177e4 1573
1da177e4
LT		 1574 /* At this point we know we have a real path component. */
1575 for(;;) {
1da177e4 1576 struct qstr this;
200e9ef7 1577 long len;
fe479a58 1578 int type;
1da177e4 1579
52094c8a 1580 err = may_lookup(nd);
1da177e4
LT		 1581 if (err)
1582 break;
1583
200e9ef7 1584 len = hash_name(name, &this.hash);
1da177e4 1585 this.name = name;
200e9ef7 1586 this.len = len;
1da177e4 1587
fe479a58 1588 type = LAST_NORM;
200e9ef7 1589 if (name[0] == '.') switch (len) {
fe479a58 1590 case 2:
200e9ef7 1591 if (name[1] == '.') {
fe479a58 1592 type = LAST_DOTDOT;
16c2cd71
AV		 1593 nd->flags |= LOOKUP_JUMPED;
1594 }
fe479a58
AV		 1595 break;
1596 case 1:
1597 type = LAST_DOT;
1598 }
5a202bcd
AV		 1599 if (likely(type == LAST_NORM)) {
1600 struct dentry *parent = nd->path.dentry;
16c2cd71 1601 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1602 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1603 err = parent->d_op->d_hash(parent, nd->inode,
1604 &this);
1605 if (err < 0)
1606 break;
1607 }
1608 }
fe479a58 1609
200e9ef7 1610 if (!name[len])
1da177e4 1611 goto last_component;
200e9ef7
LT		 1612 /*
1613 * If it wasn't NUL, we know it was '/'. Skip that
1614 * slash, and continue until no more slashes.
1615 */
1616 do {
1617 len++;
1618 } while (unlikely(name[len] == '/'));
1619 if (!name[len])
b356379a 1620 goto last_component;
200e9ef7 1621 name += len;
1da177e4 1622
ce57dfc1
AV		 1623 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1624 if (err < 0)
1625 return err;
1da177e4 1626
ce57dfc1 1627 if (err) {
b356379a 1628 err = nested_symlink(&next, nd);
1da177e4 1629 if (err)
a7472bab 1630 return err;
31e6b01f 1631 }
3ddcd056
LT		 1632 if (can_lookup(nd->inode))
1633 continue;
1da177e4 1634 err = -ENOTDIR;
3ddcd056 1635 break;
1da177e4
LT		 1636 /* here ends the main loop */
1637
1da177e4 1638last_component:
b356379a
AV		 1639 nd->last = this;
1640 nd->last_type = type;
086e183a 1641 return 0;
1da177e4 1642 }
951361f9 1643 terminate_walk(nd);
1da177e4
LT		 1644 return err;
1645}
1646
70e9b357
AV		 1647static int path_init(int dfd, const char *name, unsigned int flags,
1648 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1649{
1650 int retval = 0;
1651 int fput_needed;
1652 struct file *file;
1653
1654 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1655 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1656 nd->depth = 0;
5b6ca027
AV		 1657 if (flags & LOOKUP_ROOT) {
1658 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1659 if (*name) {
1660 if (!inode->i_op->lookup)
1661 return -ENOTDIR;
1662 retval = inode_permission(inode, MAY_EXEC);
1663 if (retval)
1664 return retval;
1665 }
5b6ca027
AV		 1666 nd->path = nd->root;
1667 nd->inode = inode;
1668 if (flags & LOOKUP_RCU) {
1669 br_read_lock(vfsmount_lock);
1670 rcu_read_lock();
1671 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1672 } else {
1673 path_get(&nd->path);
1674 }
1675 return 0;
1676 }
1677
31e6b01f 1678 nd->root.mnt = NULL;
31e6b01f
NP		 1679
1680 if (*name=='/') {
e41f7d4e
AV		 1681 if (flags & LOOKUP_RCU) {
1682 br_read_lock(vfsmount_lock);
1683 rcu_read_lock();
1684 set_root_rcu(nd);
1685 } else {
1686 set_root(nd);
1687 path_get(&nd->root);
1688 }
1689 nd->path = nd->root;
31e6b01f 1690 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1691 if (flags & LOOKUP_RCU) {
1692 struct fs_struct *fs = current->fs;
1693 unsigned seq;
31e6b01f 1694
e41f7d4e
AV		 1695 br_read_lock(vfsmount_lock);
1696 rcu_read_lock();
c28cc364 1697
e41f7d4e
AV		 1698 do {
1699 seq = read_seqcount_begin(&fs->seq);
1700 nd->path = fs->pwd;
1701 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1702 } while (read_seqcount_retry(&fs->seq, seq));
1703 } else {
1704 get_fs_pwd(current->fs, &nd->path);
1705 }
31e6b01f
NP		 1706 } else {
1707 struct dentry *dentry;
1708
1abf0c71 1709 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1710 retval = -EBADF;
1711 if (!file)
1712 goto out_fail;
1713
1714 dentry = file->f_path.dentry;
1715
f52e0c11
AV		 1716 if (*name) {
1717 retval = -ENOTDIR;
1718 if (!S_ISDIR(dentry->d_inode->i_mode))
1719 goto fput_fail;
31e6b01f 1720
4ad5abb3 1721 retval = inode_permission(dentry->d_inode, MAY_EXEC);
f52e0c11
AV		 1722 if (retval)
1723 goto fput_fail;
1724 }
31e6b01f
NP		 1725
1726 nd->path = file->f_path;
e41f7d4e
AV		 1727 if (flags & LOOKUP_RCU) {
1728 if (fput_needed)
70e9b357 1729 *fp = file;
e41f7d4e
AV		 1730 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1731 br_read_lock(vfsmount_lock);
1732 rcu_read_lock();
1733 } else {
1734 path_get(&file->f_path);
1735 fput_light(file, fput_needed);
1736 }
31e6b01f 1737 }
31e6b01f 1738
31e6b01f 1739 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1740 return 0;
2dfdd266 1741
9b4a9b14
AV		 1742fput_fail:
1743 fput_light(file, fput_needed);
1744out_fail:
1745 return retval;
1746}
1747
bd92d7fe
AV		 1748static inline int lookup_last(struct nameidata *nd, struct path *path)
1749{
1750 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1751 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1752
1753 nd->flags &= ~LOOKUP_PARENT;
1754 return walk_component(nd, path, &nd->last, nd->last_type,
1755 nd->flags & LOOKUP_FOLLOW);
1756}
1757
9b4a9b14 1758/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1759static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1760 unsigned int flags, struct nameidata *nd)
1761{
70e9b357 1762 struct file *base = NULL;
bd92d7fe
AV		 1763 struct path path;
1764 int err;
31e6b01f
NP		 1765
1766 /*
1767 * Path walking is largely split up into 2 different synchronisation
1768 * schemes, rcu-walk and ref-walk (explained in
1769 * Documentation/filesystems/path-lookup.txt). These share much of the
1770 * path walk code, but some things particularly setup, cleanup, and
1771 * following mounts are sufficiently divergent that functions are
1772 * duplicated. Typically there is a function foo(), and its RCU
1773 * analogue, foo_rcu().
1774 *
1775 * -ECHILD is the error number of choice (just to avoid clashes) that
1776 * is returned if some aspect of an rcu-walk fails. Such an error must
1777 * be handled by restarting a traditional ref-walk (which will always
1778 * be able to complete).
1779 */
bd92d7fe 1780 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1781
bd92d7fe
AV		 1782 if (unlikely(err))
1783 return err;
ee0827cd
AV		 1784
1785 current->total_link_count = 0;
bd92d7fe
AV		 1786 err = link_path_walk(name, nd);
1787
1788 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1789 err = lookup_last(nd, &path);
1790 while (err > 0) {
1791 void *cookie;
1792 struct path link = path;
bd92d7fe 1793 nd->flags |= LOOKUP_PARENT;
574197e0 1794 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1795 if (!err)
1796 err = lookup_last(nd, &path);
574197e0 1797 put_link(nd, &link, cookie);
bd92d7fe
AV		 1798 }
1799 }
ee0827cd 1800
9f1fafee
AV		 1801 if (!err)
1802 err = complete_walk(nd);
bd92d7fe
AV		 1803
1804 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1805 if (!nd->inode->i_op->lookup) {
1806 path_put(&nd->path);
bd23a539 1807 err = -ENOTDIR;
bd92d7fe
AV		 1808 }
1809 }
16c2cd71 1810
70e9b357
AV		 1811 if (base)
1812 fput(base);
ee0827cd 1813
5b6ca027 1814 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1815 path_put(&nd->root);
1816 nd->root.mnt = NULL;
1817 }
bd92d7fe 1818 return err;
ee0827cd 1819}
31e6b01f 1820
ee0827cd
AV		 1821static int do_path_lookup(int dfd, const char *name,
1822 unsigned int flags, struct nameidata *nd)
1823{
1824 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1825 if (unlikely(retval == -ECHILD))
1826 retval = path_lookupat(dfd, name, flags, nd);
1827 if (unlikely(retval == -ESTALE))
1828 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1829
1830 if (likely(!retval)) {
1831 if (unlikely(!audit_dummy_context())) {
1832 if (nd->path.dentry && nd->inode)
1833 audit_inode(name, nd->path.dentry);
1834 }
1835 }
170aa3d0 1836 return retval;
1da177e4
LT		 1837}
1838
c9c6cac0 1839int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1840{
c9c6cac0 1841 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1842}
1843
d1811465
AV		 1844int kern_path(const char *name, unsigned int flags, struct path *path)
1845{
1846 struct nameidata nd;
1847 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1848 if (!res)
1849 *path = nd.path;
1850 return res;
1851}
1852
16f18200
JJS		 1853/**
1854 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1855 * @dentry: pointer to dentry of the base directory
1856 * @mnt: pointer to vfs mount of the base directory
1857 * @name: pointer to file name
1858 * @flags: lookup flags
e0a01249 1859 * @path: pointer to struct path to fill
16f18200
JJS		 1860 */
1861int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1862 const char *name, unsigned int flags,
e0a01249 1863 struct path *path)
16f18200 1864{
e0a01249
AV		 1865 struct nameidata nd;
1866 int err;
1867 nd.root.dentry = dentry;
1868 nd.root.mnt = mnt;
1869 BUG_ON(flags & LOOKUP_PARENT);
5b6ca027 1870 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
e0a01249
AV		 1871 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1872 if (!err)
1873 *path = nd.path;
1874 return err;
16f18200
JJS		 1875}
1876
057f6c01
JM		 1877/*
1878 * Restricted form of lookup. Doesn't follow links, single-component only,
1879 * needs parent already locked. Doesn't follow mounts.
1880 * SMP-safe.
1881 */
eead1911 1882static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1883{
4ac91378 1884 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1885}
1886
eead1911 1887/**
a6b91919 1888 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1889 * @name: pathname component to lookup
1890 * @base: base directory to lookup from
1891 * @len: maximum length @len should be interpreted to
1892 *
a6b91919
RD		 1893 * Note that this routine is purely a helper for filesystem usage and should
1894 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1895 * nameidata argument is passed to the filesystem methods and a filesystem
1896 * using this helper needs to be prepared for that.
1897 */
057f6c01
JM		 1898struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1899{
057f6c01 1900 struct qstr this;
6a96ba54 1901 unsigned int c;
cda309de 1902 int err;
057f6c01 1903
2f9092e1
DW		 1904 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1905
6a96ba54
AV		 1906 this.name = name;
1907 this.len = len;
0145acc2 1908 this.hash = full_name_hash(name, len);
6a96ba54
AV		 1909 if (!len)
1910 return ERR_PTR(-EACCES);
1911
6a96ba54
AV		 1912 while (len--) {
1913 c = *(const unsigned char *)name++;
1914 if (c == '/' || c == '\0')
1915 return ERR_PTR(-EACCES);
6a96ba54 1916 }
5a202bcd
AV		 1917 /*
1918 * See if the low-level filesystem might want
1919 * to use its own hash..
1920 */
1921 if (base->d_flags & DCACHE_OP_HASH) {
1922 int err = base->d_op->d_hash(base, base->d_inode, &this);
1923 if (err < 0)
1924 return ERR_PTR(err);
1925 }
eead1911 1926
cda309de
MS		 1927 err = inode_permission(base->d_inode, MAY_EXEC);
1928 if (err)
1929 return ERR_PTR(err);
1930
49705b77 1931 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1932}
1933
1fa1e7f6
AW		 1934int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1935 struct path *path, int *empty)
1da177e4 1936{
2d8f3038 1937 struct nameidata nd;
1fa1e7f6 1938 char *tmp = getname_flags(name, flags, empty);
1da177e4 1939 int err = PTR_ERR(tmp);
1da177e4 1940 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1941
1942 BUG_ON(flags & LOOKUP_PARENT);
1943
1944 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1945 putname(tmp);
2d8f3038
AV		 1946 if (!err)
1947 *path = nd.path;
1da177e4
LT		 1948 }
1949 return err;
1950}
1951
1fa1e7f6
AW		 1952int user_path_at(int dfd, const char __user *name, unsigned flags,
1953 struct path *path)
1954{
f7493e5d 1955 return user_path_at_empty(dfd, name, flags, path, NULL);
1fa1e7f6
AW		 1956}
1957
2ad94ae6
AV		 1958static int user_path_parent(int dfd, const char __user *path,
1959 struct nameidata *nd, char **name)
1960{
1961 char *s = getname(path);
1962 int error;
1963
1964 if (IS_ERR(s))
1965 return PTR_ERR(s);
1966
1967 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1968 if (error)
1969 putname(s);
1970 else
1971 *name = s;
1972
1973 return error;
1974}
1975
1da177e4
LT		 1976/*
1977 * It's inline, so penalty for filesystems that don't use sticky bit is
1978 * minimal.
1979 */
1980static inline int check_sticky(struct inode *dir, struct inode *inode)
1981{
da9592ed
DH		 1982 uid_t fsuid = current_fsuid();
1983
1da177e4
LT		 1984 if (!(dir->i_mode & S_ISVTX))
1985 return 0;
e795b717
SH		 1986 if (current_user_ns() != inode_userns(inode))
1987 goto other_userns;
da9592ed 1988 if (inode->i_uid == fsuid)
1da177e4 1989 return 0;
da9592ed 1990 if (dir->i_uid == fsuid)
1da177e4 1991 return 0;
e795b717
SH		 1992
1993other_userns:
1994 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1da177e4
LT		 1995}
1996
1997/*
1998 * Check whether we can remove a link victim from directory dir, check
1999 * whether the type of victim is right.
2000 * 1. We can't do it if dir is read-only (done in permission())
2001 * 2. We should have write and exec permissions on dir
2002 * 3. We can't remove anything from append-only dir
2003 * 4. We can't do anything with immutable dir (done in permission())
2004 * 5. If the sticky bit on dir is set we should either
2005 * a. be owner of dir, or
2006 * b. be owner of victim, or
2007 * c. have CAP_FOWNER capability
2008 * 6. If the victim is append-only or immutable we can't do antyhing with
2009 * links pointing to it.
2010 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2011 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2012 * 9. We can't remove a root or mountpoint.
2013 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2014 * nfs_async_unlink().
2015 */
858119e1 2016static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 2017{
2018 int error;
2019
2020 if (!victim->d_inode)
2021 return -ENOENT;
2022
2023 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 2024 audit_inode_child(victim, dir);
1da177e4 2025
f419a2e3 2026 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 2027 if (error)
2028 return error;
2029 if (IS_APPEND(dir))
2030 return -EPERM;
2031 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 2032 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 2033 return -EPERM;
2034 if (isdir) {
2035 if (!S_ISDIR(victim->d_inode->i_mode))
2036 return -ENOTDIR;
2037 if (IS_ROOT(victim))
2038 return -EBUSY;
2039 } else if (S_ISDIR(victim->d_inode->i_mode))
2040 return -EISDIR;
2041 if (IS_DEADDIR(dir))
2042 return -ENOENT;
2043 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2044 return -EBUSY;
2045 return 0;
2046}
2047
2048/* Check whether we can create an object with dentry child in directory
2049 * dir.
2050 * 1. We can't do it if child already exists (open has special treatment for
2051 * this case, but since we are inlined it's OK)
2052 * 2. We can't do it if dir is read-only (done in permission())
2053 * 3. We should have write and exec permissions on dir
2054 * 4. We can't do it if dir is immutable (done in permission())
2055 */
a95164d9 2056static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 2057{
2058 if (child->d_inode)
2059 return -EEXIST;
2060 if (IS_DEADDIR(dir))
2061 return -ENOENT;
f419a2e3 2062 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 2063}
2064
1da177e4
LT		 2065/*
2066 * p1 and p2 should be directories on the same fs.
2067 */
2068struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2069{
2070 struct dentry *p;
2071
2072 if (p1 == p2) {
f2eace23 2073 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 2074 return NULL;
2075 }
2076
a11f3a05 2077 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 2078
e2761a11
OH		 2079 p = d_ancestor(p2, p1);
2080 if (p) {
2081 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2082 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2083 return p;
1da177e4
LT		 2084 }
2085
e2761a11
OH		 2086 p = d_ancestor(p1, p2);
2087 if (p) {
2088 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2089 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2090 return p;
1da177e4
LT		 2091 }
2092
f2eace23
IM		 2093 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2094 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 2095 return NULL;
2096}
2097
2098void unlock_rename(struct dentry *p1, struct dentry *p2)
2099{
1b1dcc1b 2100 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 2101 if (p1 != p2) {
1b1dcc1b 2102 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 2103 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 2104 }
2105}
2106
4acdaf27 2107int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1da177e4
LT		 2108 struct nameidata *nd)
2109{
a95164d9 2110 int error = may_create(dir, dentry);
1da177e4
LT		 2111
2112 if (error)
2113 return error;
2114
acfa4380 2115 if (!dir->i_op->create)
1da177e4
LT		 2116 return -EACCES; /* shouldn't it be ENOSYS? */
2117 mode &= S_IALLUGO;
2118 mode |= S_IFREG;
2119 error = security_inode_create(dir, dentry, mode);
2120 if (error)
2121 return error;
1da177e4 2122 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 2123 if (!error)
f38aa942 2124 fsnotify_create(dir, dentry);
1da177e4
LT		 2125 return error;
2126}
2127
73d049a4 2128static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 2129{
3fb64190 2130 struct dentry *dentry = path->dentry;
1da177e4
LT		 2131 struct inode *inode = dentry->d_inode;
2132 int error;
2133
bcda7652
AV		 2134 /* O_PATH? */
2135 if (!acc_mode)
2136 return 0;
2137
1da177e4
LT		 2138 if (!inode)
2139 return -ENOENT;
2140
c8fe8f30
CH		 2141 switch (inode->i_mode & S_IFMT) {
2142 case S_IFLNK:
1da177e4 2143 return -ELOOP;
c8fe8f30
CH		 2144 case S_IFDIR:
2145 if (acc_mode & MAY_WRITE)
2146 return -EISDIR;
2147 break;
2148 case S_IFBLK:
2149 case S_IFCHR:
3fb64190 2150 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 2151 return -EACCES;
c8fe8f30
CH		 2152 /*FALLTHRU*/
2153 case S_IFIFO:
2154 case S_IFSOCK:
1da177e4 2155 flag &= ~O_TRUNC;
c8fe8f30 2156 break;
4a3fd211 2157 }
b41572e9 2158
3fb64190 2159 error = inode_permission(inode, acc_mode);
b41572e9
DH		 2160 if (error)
2161 return error;
6146f0d5 2162
1da177e4
LT		 2163 /*
2164 * An append-only file must be opened in append mode for writing.
2165 */
2166 if (IS_APPEND(inode)) {
8737c930 2167 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 2168 return -EPERM;
1da177e4 2169 if (flag & O_TRUNC)
7715b521 2170 return -EPERM;
1da177e4
LT		 2171 }
2172
2173 /* O_NOATIME can only be set by the owner or superuser */
2e149670 2174 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 2175 return -EPERM;
1da177e4 2176
f3c7691e 2177 return 0;
7715b521 2178}
1da177e4 2179
e1181ee6 2180static int handle_truncate(struct file *filp)
7715b521 2181{
e1181ee6 2182 struct path *path = &filp->f_path;
7715b521
AV		 2183 struct inode *inode = path->dentry->d_inode;
2184 int error = get_write_access(inode);
2185 if (error)
2186 return error;
2187 /*
2188 * Refuse to truncate files with mandatory locks held on them.
2189 */
2190 error = locks_verify_locked(inode);
2191 if (!error)
ea0d3ab2 2192 error = security_path_truncate(path);
7715b521
AV		 2193 if (!error) {
2194 error = do_truncate(path->dentry, 0,
2195 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2196 filp);
7715b521
AV		 2197 }
2198 put_write_access(inode);
acd0c935 2199 return error;
1da177e4
LT		 2200}
2201
d57999e1
DH		 2202static inline int open_to_namei_flags(int flag)
2203{
8a5e929d
AV		 2204 if ((flag & O_ACCMODE) == 3)
2205 flag--;
d57999e1
DH		 2206 return flag;
2207}
2208
31e6b01f 2209/*
fe2d35ff 2210 * Handle the last step of open()
31e6b01f 2211 */
fb1cc555 2212static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2213 const struct open_flags *op, const char *pathname)
fb1cc555 2214{
a1e28038 2215 struct dentry *dir = nd->path.dentry;
6c0d46c4 2216 struct dentry *dentry;
ca344a89 2217 int open_flag = op->open_flag;
6c0d46c4 2218 int will_truncate = open_flag & O_TRUNC;
ca344a89 2219 int want_write = 0;
bcda7652 2220 int acc_mode = op->acc_mode;
fb1cc555 2221 struct file *filp;
16c2cd71 2222 int error;
1f36f774 2223
c3e380b0
AV		 2224 nd->flags &= ~LOOKUP_PARENT;
2225 nd->flags |= op->intent;
2226
1f36f774
AV		 2227 switch (nd->last_type) {
2228 case LAST_DOTDOT:
176306f5 2229 case LAST_DOT:
fe2d35ff
AV		 2230 error = handle_dots(nd, nd->last_type);
2231 if (error)
2232 return ERR_PTR(error);
1f36f774 2233 /* fallthrough */
1f36f774 2234 case LAST_ROOT:
9f1fafee 2235 error = complete_walk(nd);
16c2cd71 2236 if (error)
9f1fafee 2237 return ERR_PTR(error);
fe2d35ff 2238 audit_inode(pathname, nd->path.dentry);
ca344a89 2239 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2240 error = -EISDIR;
2241 goto exit;
2242 }
2243 goto ok;
1f36f774 2244 case LAST_BIND:
9f1fafee 2245 error = complete_walk(nd);
16c2cd71 2246 if (error)
9f1fafee 2247 return ERR_PTR(error);
1f36f774 2248 audit_inode(pathname, dir);
67ee3ad2 2249 goto ok;
1f36f774 2250 }
67ee3ad2 2251
ca344a89 2252 if (!(open_flag & O_CREAT)) {
bcda7652 2253 int symlink_ok = 0;
fe2d35ff
AV		 2254 if (nd->last.name[nd->last.len])
2255 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2256 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2257 symlink_ok = 1;
fe2d35ff 2258 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2259 error = walk_component(nd, path, &nd->last, LAST_NORM,
2260 !symlink_ok);
2261 if (error < 0)
fe2d35ff 2262 return ERR_PTR(error);
ce57dfc1 2263 if (error) /* symlink */
fe2d35ff 2264 return NULL;
fe2d35ff 2265 /* sayonara */
9f1fafee
AV		 2266 error = complete_walk(nd);
2267 if (error)
7f6c7e62 2268 return ERR_PTR(error);
fe2d35ff
AV		 2269
2270 error = -ENOTDIR;
2271 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2272 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2273 goto exit;
2274 }
2275 audit_inode(pathname, nd->path.dentry);
2276 goto ok;
2277 }
2278
2279 /* create side of things */
a3fbbde7
AV		 2280 /*
2281 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2282 * cleared when we got to the last component we are about to look up
2283 */
9f1fafee
AV		 2284 error = complete_walk(nd);
2285 if (error)
2286 return ERR_PTR(error);
fe2d35ff
AV		 2287
2288 audit_inode(pathname, dir);
16c2cd71 2289 error = -EISDIR;
1f36f774 2290 /* trailing slashes? */
31e6b01f
NP		 2291 if (nd->last.name[nd->last.len])
2292 goto exit;
a2c36b45 2293
a1e28038
AV		 2294 mutex_lock(&dir->d_inode->i_mutex);
2295
6c0d46c4
AV		 2296 dentry = lookup_hash(nd);
2297 error = PTR_ERR(dentry);
2298 if (IS_ERR(dentry)) {
fb1cc555
AV		 2299 mutex_unlock(&dir->d_inode->i_mutex);
2300 goto exit;
2301 }
2302
6c0d46c4
AV		 2303 path->dentry = dentry;
2304 path->mnt = nd->path.mnt;
2305
fb1cc555 2306 /* Negative dentry, just create the file */
6c0d46c4 2307 if (!dentry->d_inode) {
a218d0fd 2308 umode_t mode = op->mode;
6c0d46c4
AV		 2309 if (!IS_POSIXACL(dir->d_inode))
2310 mode &= ~current_umask();
fb1cc555
AV		 2311 /*
2312 * This write is needed to ensure that a
6c0d46c4 2313 * rw->ro transition does not occur between
fb1cc555
AV		 2314 * the time when the file is created and when
2315 * a permanent write count is taken through
2316 * the 'struct file' in nameidata_to_filp().
2317 */
2318 error = mnt_want_write(nd->path.mnt);
2319 if (error)
2320 goto exit_mutex_unlock;
ca344a89 2321 want_write = 1;
9b44f1b3 2322 /* Don't check for write permission, don't truncate */
ca344a89 2323 open_flag &= ~O_TRUNC;
6c0d46c4 2324 will_truncate = 0;
bcda7652 2325 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2326 error = security_path_mknod(&nd->path, dentry, mode, 0);
2327 if (error)
2328 goto exit_mutex_unlock;
2329 error = vfs_create(dir->d_inode, dentry, mode, nd);
2330 if (error)
2331 goto exit_mutex_unlock;
2332 mutex_unlock(&dir->d_inode->i_mutex);
2333 dput(nd->path.dentry);
2334 nd->path.dentry = dentry;
ca344a89 2335 goto common;
fb1cc555
AV		 2336 }
2337
2338 /*
2339 * It already exists.
2340 */
2341 mutex_unlock(&dir->d_inode->i_mutex);
2342 audit_inode(pathname, path->dentry);
2343
2344 error = -EEXIST;
ca344a89 2345 if (open_flag & O_EXCL)
fb1cc555
AV		 2346 goto exit_dput;
2347
9875cf80
DH		 2348 error = follow_managed(path, nd->flags);
2349 if (error < 0)
2350 goto exit_dput;
fb1cc555 2351
a3fbbde7
AV		 2352 if (error)
2353 nd->flags |= LOOKUP_JUMPED;
2354
fb1cc555
AV		 2355 error = -ENOENT;
2356 if (!path->dentry->d_inode)
2357 goto exit_dput;
9e67f361
AV		 2358
2359 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2360 return NULL;
fb1cc555
AV		 2361
2362 path_to_nameidata(path, nd);
31e6b01f 2363 nd->inode = path->dentry->d_inode;
a3fbbde7
AV		 2364 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2365 error = complete_walk(nd);
2366 if (error)
097b180c 2367 return ERR_PTR(error);
fb1cc555 2368 error = -EISDIR;
31e6b01f 2369 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2370 goto exit;
67ee3ad2 2371ok:
6c0d46c4
AV		 2372 if (!S_ISREG(nd->inode->i_mode))
2373 will_truncate = 0;
2374
0f9d1a10
AV		 2375 if (will_truncate) {
2376 error = mnt_want_write(nd->path.mnt);
2377 if (error)
2378 goto exit;
ca344a89 2379 want_write = 1;
0f9d1a10 2380 }
ca344a89 2381common:
bcda7652 2382 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2383 if (error)
0f9d1a10 2384 goto exit;
0f9d1a10
AV		 2385 filp = nameidata_to_filp(nd);
2386 if (!IS_ERR(filp)) {
2387 error = ima_file_check(filp, op->acc_mode);
2388 if (error) {
2389 fput(filp);
2390 filp = ERR_PTR(error);
2391 }
2392 }
2393 if (!IS_ERR(filp)) {
2394 if (will_truncate) {
2395 error = handle_truncate(filp);
2396 if (error) {
2397 fput(filp);
2398 filp = ERR_PTR(error);
2399 }
2400 }
2401 }
ca344a89
AV		 2402out:
2403 if (want_write)
0f9d1a10
AV		 2404 mnt_drop_write(nd->path.mnt);
2405 path_put(&nd->path);
fb1cc555
AV		 2406 return filp;
2407
2408exit_mutex_unlock:
2409 mutex_unlock(&dir->d_inode->i_mutex);
2410exit_dput:
2411 path_put_conditional(path, nd);
2412exit:
ca344a89
AV		 2413 filp = ERR_PTR(error);
2414 goto out;
fb1cc555
AV		 2415}
2416
13aab428 2417static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2418 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2419{
fe2d35ff 2420 struct file *base = NULL;
4a3fd211 2421 struct file *filp;
9850c056 2422 struct path path;
13aab428 2423 int error;
31e6b01f
NP		 2424
2425 filp = get_empty_filp();
2426 if (!filp)
2427 return ERR_PTR(-ENFILE);
2428
47c805dc 2429 filp->f_flags = op->open_flag;
73d049a4
AV		 2430 nd->intent.open.file = filp;
2431 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2432 nd->intent.open.create_mode = op->mode;
31e6b01f 2433
73d049a4 2434 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2435 if (unlikely(error))
13aab428 2436 goto out_filp;
31e6b01f 2437
fe2d35ff 2438 current->total_link_count = 0;
73d049a4 2439 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2440 if (unlikely(error))
2441 goto out_filp;
1da177e4 2442
73d049a4 2443 filp = do_last(nd, &path, op, pathname);
806b681c 2444 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2445 struct path link = path;
def4af30 2446 void *cookie;
574197e0 2447 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2448 path_put_conditional(&path, nd);
2449 path_put(&nd->path);
40b39136
AV		 2450 filp = ERR_PTR(-ELOOP);
2451 break;
2452 }
73d049a4
AV		 2453 nd->flags |= LOOKUP_PARENT;
2454 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2455 error = follow_link(&link, nd, &cookie);
c3e380b0 2456 if (unlikely(error))
f1afe9ef 2457 filp = ERR_PTR(error);
c3e380b0 2458 else
73d049a4 2459 filp = do_last(nd, &path, op, pathname);
574197e0 2460 put_link(nd, &link, cookie);
806b681c 2461 }
10fa8e62 2462out:
73d049a4
AV		 2463 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2464 path_put(&nd->root);
fe2d35ff
AV		 2465 if (base)
2466 fput(base);
73d049a4 2467 release_open_intent(nd);
10fa8e62 2468 return filp;
1da177e4 2469
31e6b01f 2470out_filp:
806b681c 2471 filp = ERR_PTR(error);
10fa8e62 2472 goto out;
1da177e4
LT		 2473}
2474
13aab428
AV		 2475struct file *do_filp_open(int dfd, const char *pathname,
2476 const struct open_flags *op, int flags)
2477{
73d049a4 2478 struct nameidata nd;
13aab428
AV		 2479 struct file *filp;
2480
73d049a4 2481 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2482 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2483 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2484 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2485 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2486 return filp;
2487}
2488
73d049a4
AV		 2489struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2490 const char *name, const struct open_flags *op, int flags)
2491{
2492 struct nameidata nd;
2493 struct file *file;
2494
2495 nd.root.mnt = mnt;
2496 nd.root.dentry = dentry;
2497
2498 flags |= LOOKUP_ROOT;
2499
bcda7652 2500 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2501 return ERR_PTR(-ELOOP);
2502
2503 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2504 if (unlikely(file == ERR_PTR(-ECHILD)))
2505 file = path_openat(-1, name, &nd, op, flags);
2506 if (unlikely(file == ERR_PTR(-ESTALE)))
2507 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2508 return file;
2509}
2510
ed75e95d 2511struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
1da177e4 2512{
c663e5d8 2513 struct dentry *dentry = ERR_PTR(-EEXIST);
ed75e95d
AV		 2514 struct nameidata nd;
2515 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2516 if (error)
2517 return ERR_PTR(error);
1da177e4 2518
c663e5d8
CH		 2519 /*
2520 * Yucky last component or no last component at all?
2521 * (foo/., foo/.., /////)
2522 */
ed75e95d
AV		 2523 if (nd.last_type != LAST_NORM)
2524 goto out;
2525 nd.flags &= ~LOOKUP_PARENT;
2526 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2527 nd.intent.open.flags = O_EXCL;
c663e5d8
CH		 2528
2529 /*
2530 * Do the final lookup.
2531 */
ed75e95d
AV		 2532 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2533 dentry = lookup_hash(&nd);
1da177e4
LT		 2534 if (IS_ERR(dentry))
2535 goto fail;
c663e5d8 2536
e9baf6e5
AV		 2537 if (dentry->d_inode)
2538 goto eexist;
c663e5d8
CH		 2539 /*
2540 * Special case - lookup gave negative, but... we had foo/bar/
2541 * From the vfs_mknod() POV we just have a negative dentry -
2542 * all is fine. Let's be bastards - you had / on the end, you've
2543 * been asking for (non-existent) directory. -ENOENT for you.
2544 */
ed75e95d 2545 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
e9baf6e5
AV		 2546 dput(dentry);
2547 dentry = ERR_PTR(-ENOENT);
ed75e95d 2548 goto fail;
e9baf6e5 2549 }
ed75e95d 2550 *path = nd.path;
1da177e4 2551 return dentry;
e9baf6e5 2552eexist:
1da177e4 2553 dput(dentry);
e9baf6e5 2554 dentry = ERR_PTR(-EEXIST);
1da177e4 2555fail:
ed75e95d
AV		 2556 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2557out:
2558 path_put(&nd.path);
1da177e4
LT		 2559 return dentry;
2560}
dae6ad8f
AV		 2561EXPORT_SYMBOL(kern_path_create);
2562
2563struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2564{
2565 char *tmp = getname(pathname);
2566 struct dentry *res;
2567 if (IS_ERR(tmp))
2568 return ERR_CAST(tmp);
2569 res = kern_path_create(dfd, tmp, path, is_dir);
2570 putname(tmp);
2571 return res;
2572}
2573EXPORT_SYMBOL(user_path_create);
2574
1a67aafb 2575int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1da177e4 2576{
a95164d9 2577 int error = may_create(dir, dentry);
1da177e4
LT		 2578
2579 if (error)
2580 return error;
2581
e795b717
SH		 2582 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2583 !ns_capable(inode_userns(dir), CAP_MKNOD))
1da177e4
LT		 2584 return -EPERM;
2585
acfa4380 2586 if (!dir->i_op->mknod)
1da177e4
LT		 2587 return -EPERM;
2588
08ce5f16
SH		 2589 error = devcgroup_inode_mknod(mode, dev);
2590 if (error)
2591 return error;
2592
1da177e4
LT		 2593 error = security_inode_mknod(dir, dentry, mode, dev);
2594 if (error)
2595 return error;
2596
1da177e4 2597 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2598 if (!error)
f38aa942 2599 fsnotify_create(dir, dentry);
1da177e4
LT		 2600 return error;
2601}
2602
f69aac00 2603static int may_mknod(umode_t mode)
463c3197
DH		 2604{
2605 switch (mode & S_IFMT) {
2606 case S_IFREG:
2607 case S_IFCHR:
2608 case S_IFBLK:
2609 case S_IFIFO:
2610 case S_IFSOCK:
2611 case 0: /* zero mode translates to S_IFREG */
2612 return 0;
2613 case S_IFDIR:
2614 return -EPERM;
2615 default:
2616 return -EINVAL;
2617 }
2618}
2619
8208a22b 2620SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2e4d0924 2621 unsigned, dev)
1da177e4 2622{
2ad94ae6 2623 struct dentry *dentry;
dae6ad8f
AV		 2624 struct path path;
2625 int error;
1da177e4
LT		 2626
2627 if (S_ISDIR(mode))
2628 return -EPERM;
1da177e4 2629
dae6ad8f
AV		 2630 dentry = user_path_create(dfd, filename, &path, 0);
2631 if (IS_ERR(dentry))
2632 return PTR_ERR(dentry);
2ad94ae6 2633
dae6ad8f 2634 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2635 mode &= ~current_umask();
463c3197
DH		 2636 error = may_mknod(mode);
2637 if (error)
2638 goto out_dput;
dae6ad8f 2639 error = mnt_want_write(path.mnt);
463c3197
DH		 2640 if (error)
2641 goto out_dput;
dae6ad8f 2642 error = security_path_mknod(&path, dentry, mode, dev);
be6d3e56
KT		 2643 if (error)
2644 goto out_drop_write;
463c3197 2645 switch (mode & S_IFMT) {
1da177e4 2646 case 0: case S_IFREG:
dae6ad8f 2647 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
1da177e4
LT		 2648 break;
2649 case S_IFCHR: case S_IFBLK:
dae6ad8f 2650 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2651 new_decode_dev(dev));
2652 break;
2653 case S_IFIFO: case S_IFSOCK:
dae6ad8f 2654 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
1da177e4 2655 break;
1da177e4 2656 }
be6d3e56 2657out_drop_write:
dae6ad8f 2658 mnt_drop_write(path.mnt);
463c3197
DH		 2659out_dput:
2660 dput(dentry);
dae6ad8f
AV		 2661 mutex_unlock(&path.dentry->d_inode->i_mutex);
2662 path_put(&path);
1da177e4
LT		 2663
2664 return error;
2665}
2666
8208a22b 2667SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
5590ff0d
UD		 2668{
2669 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2670}
2671
18bb1db3 2672int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1da177e4 2673{
a95164d9 2674 int error = may_create(dir, dentry);
8de52778 2675 unsigned max_links = dir->i_sb->s_max_links;
1da177e4
LT		 2676
2677 if (error)
2678 return error;
2679
acfa4380 2680 if (!dir->i_op->mkdir)
1da177e4
LT		 2681 return -EPERM;
2682
2683 mode &= (S_IRWXUGO|S_ISVTX);
2684 error = security_inode_mkdir(dir, dentry, mode);
2685 if (error)
2686 return error;
2687
8de52778
AV		 2688 if (max_links && dir->i_nlink >= max_links)
2689 return -EMLINK;
2690
1da177e4 2691 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2692 if (!error)
f38aa942 2693 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2694 return error;
2695}
2696
a218d0fd 2697SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
1da177e4 2698{
6902d925 2699 struct dentry *dentry;
dae6ad8f
AV		 2700 struct path path;
2701 int error;
1da177e4 2702
dae6ad8f 2703 dentry = user_path_create(dfd, pathname, &path, 1);
6902d925 2704 if (IS_ERR(dentry))
dae6ad8f 2705 return PTR_ERR(dentry);
1da177e4 2706
dae6ad8f 2707 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2708 mode &= ~current_umask();
dae6ad8f 2709 error = mnt_want_write(path.mnt);
463c3197
DH		 2710 if (error)
2711 goto out_dput;
dae6ad8f 2712 error = security_path_mkdir(&path, dentry, mode);
be6d3e56
KT		 2713 if (error)
2714 goto out_drop_write;
dae6ad8f 2715 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
be6d3e56 2716out_drop_write:
dae6ad8f 2717 mnt_drop_write(path.mnt);
463c3197 2718out_dput:
6902d925 2719 dput(dentry);
dae6ad8f
AV		 2720 mutex_unlock(&path.dentry->d_inode->i_mutex);
2721 path_put(&path);
1da177e4
LT		 2722 return error;
2723}
2724
a218d0fd 2725SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
5590ff0d
UD		 2726{
2727 return sys_mkdirat(AT_FDCWD, pathname, mode);
2728}
2729
1da177e4 2730/*
a71905f0
SW		 2731 * The dentry_unhash() helper will try to drop the dentry early: we
2732 * should have a usage count of 2 if we're the only user of this
2733 * dentry, and if that is true (possibly after pruning the dcache),
2734 * then we drop the dentry now.
1da177e4
LT		 2735 *
2736 * A low-level filesystem can, if it choses, legally
2737 * do a
2738 *
2739 * if (!d_unhashed(dentry))
2740 * return -EBUSY;
2741 *
2742 * if it cannot handle the case of removing a directory
2743 * that is still in use by something else..
2744 */
2745void dentry_unhash(struct dentry *dentry)
2746{
dc168427 2747 shrink_dcache_parent(dentry);
1da177e4 2748 spin_lock(&dentry->d_lock);
64252c75 2749 if (dentry->d_count == 1)
1da177e4
LT		 2750 __d_drop(dentry);
2751 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2752}
2753
2754int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2755{
2756 int error = may_delete(dir, dentry, 1);
2757
2758 if (error)
2759 return error;
2760
acfa4380 2761 if (!dir->i_op->rmdir)
1da177e4
LT		 2762 return -EPERM;
2763
1d2ef590 2764 dget(dentry);
1b1dcc1b 2765 mutex_lock(&dentry->d_inode->i_mutex);
912dbc15
SW		 2766
2767 error = -EBUSY;
1da177e4 2768 if (d_mountpoint(dentry))
912dbc15
SW		 2769 goto out;
2770
2771 error = security_inode_rmdir(dir, dentry);
2772 if (error)
2773 goto out;
2774
3cebde24 2775 shrink_dcache_parent(dentry);
912dbc15
SW		 2776 error = dir->i_op->rmdir(dir, dentry);
2777 if (error)
2778 goto out;
2779
2780 dentry->d_inode->i_flags |= S_DEAD;
2781 dont_mount(dentry);
2782
2783out:
1b1dcc1b 2784 mutex_unlock(&dentry->d_inode->i_mutex);
1d2ef590 2785 dput(dentry);
912dbc15 2786 if (!error)
1da177e4 2787 d_delete(dentry);
1da177e4
LT		 2788 return error;
2789}
2790
5590ff0d 2791static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2792{
2793 int error = 0;
2794 char * name;
2795 struct dentry *dentry;
2796 struct nameidata nd;
2797
2ad94ae6 2798 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2799 if (error)
2ad94ae6 2800 return error;
1da177e4
LT		 2801
2802 switch(nd.last_type) {
0612d9fb
OH		 2803 case LAST_DOTDOT:
2804 error = -ENOTEMPTY;
2805 goto exit1;
2806 case LAST_DOT:
2807 error = -EINVAL;
2808 goto exit1;
2809 case LAST_ROOT:
2810 error = -EBUSY;
2811 goto exit1;
1da177e4 2812 }
0612d9fb
OH		 2813
2814 nd.flags &= ~LOOKUP_PARENT;
2815
4ac91378 2816 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2817 dentry = lookup_hash(&nd);
1da177e4 2818 error = PTR_ERR(dentry);
6902d925
DH		 2819 if (IS_ERR(dentry))
2820 goto exit2;
e6bc45d6
TT		 2821 if (!dentry->d_inode) {
2822 error = -ENOENT;
2823 goto exit3;
2824 }
0622753b
DH		 2825 error = mnt_want_write(nd.path.mnt);
2826 if (error)
2827 goto exit3;
be6d3e56
KT		 2828 error = security_path_rmdir(&nd.path, dentry);
2829 if (error)
2830 goto exit4;
4ac91378 2831 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2832exit4:
0622753b
DH		 2833 mnt_drop_write(nd.path.mnt);
2834exit3:
6902d925
DH		 2835 dput(dentry);
2836exit2:
4ac91378 2837 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2838exit1:
1d957f9b 2839 path_put(&nd.path);
1da177e4
LT		 2840 putname(name);
2841 return error;
2842}
2843
3cdad428 2844SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2845{
2846 return do_rmdir(AT_FDCWD, pathname);
2847}
2848
1da177e4
LT		 2849int vfs_unlink(struct inode *dir, struct dentry *dentry)
2850{
2851 int error = may_delete(dir, dentry, 0);
2852
2853 if (error)
2854 return error;
2855
acfa4380 2856 if (!dir->i_op->unlink)
1da177e4
LT		 2857 return -EPERM;
2858
1b1dcc1b 2859 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2860 if (d_mountpoint(dentry))
2861 error = -EBUSY;
2862 else {
2863 error = security_inode_unlink(dir, dentry);
bec1052e 2864 if (!error) {
1da177e4 2865 error = dir->i_op->unlink(dir, dentry);
bec1052e 2866 if (!error)
d83c49f3 2867 dont_mount(dentry);
bec1052e 2868 }
1da177e4 2869 }
1b1dcc1b 2870 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2871
2872 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2873 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2874 fsnotify_link_count(dentry->d_inode);
e234f35c 2875 d_delete(dentry);
1da177e4 2876 }
0eeca283 2877
1da177e4
LT		 2878 return error;
2879}
2880
2881/*
2882 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2883 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2884 * writeout happening, and we don't want to prevent access to the directory
2885 * while waiting on the I/O.
2886 */
5590ff0d 2887static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2888{
2ad94ae6
AV		 2889 int error;
2890 char *name;
1da177e4
LT		 2891 struct dentry *dentry;
2892 struct nameidata nd;
2893 struct inode *inode = NULL;
2894
2ad94ae6 2895 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2896 if (error)
2ad94ae6
AV		 2897 return error;
2898
1da177e4
LT		 2899 error = -EISDIR;
2900 if (nd.last_type != LAST_NORM)
2901 goto exit1;
0612d9fb
OH		 2902
2903 nd.flags &= ~LOOKUP_PARENT;
2904
4ac91378 2905 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2906 dentry = lookup_hash(&nd);
1da177e4
LT		 2907 error = PTR_ERR(dentry);
2908 if (!IS_ERR(dentry)) {
2909 /* Why not before? Because we want correct error value */
50338b88
TE		 2910 if (nd.last.name[nd.last.len])
2911 goto slashes;
1da177e4 2912 inode = dentry->d_inode;
50338b88 2913 if (!inode)
e6bc45d6
TT		 2914 goto slashes;
2915 ihold(inode);
0622753b
DH		 2916 error = mnt_want_write(nd.path.mnt);
2917 if (error)
2918 goto exit2;
be6d3e56
KT		 2919 error = security_path_unlink(&nd.path, dentry);
2920 if (error)
2921 goto exit3;
4ac91378 2922 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2923exit3:
0622753b 2924 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2925 exit2:
2926 dput(dentry);
2927 }
4ac91378 2928 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2929 if (inode)
2930 iput(inode); /* truncate the inode here */
2931exit1:
1d957f9b 2932 path_put(&nd.path);
1da177e4
LT		 2933 putname(name);
2934 return error;
2935
2936slashes:
2937 error = !dentry->d_inode ? -ENOENT :
2938 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2939 goto exit2;
2940}
2941
2e4d0924 2942SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2943{
2944 if ((flag & ~AT_REMOVEDIR) != 0)
2945 return -EINVAL;
2946
2947 if (flag & AT_REMOVEDIR)
2948 return do_rmdir(dfd, pathname);
2949
2950 return do_unlinkat(dfd, pathname);
2951}
2952
3480b257 2953SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2954{
2955 return do_unlinkat(AT_FDCWD, pathname);
2956}
2957
db2e747b 2958int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2959{
a95164d9 2960 int error = may_create(dir, dentry);
1da177e4
LT		 2961
2962 if (error)
2963 return error;
2964
acfa4380 2965 if (!dir->i_op->symlink)
1da177e4
LT		 2966 return -EPERM;
2967
2968 error = security_inode_symlink(dir, dentry, oldname);
2969 if (error)
2970 return error;
2971
1da177e4 2972 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2973 if (!error)
f38aa942 2974 fsnotify_create(dir, dentry);
1da177e4
LT		 2975 return error;
2976}
2977
2e4d0924
HC		 2978SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2979 int, newdfd, const char __user *, newname)
1da177e4 2980{
2ad94ae6
AV		 2981 int error;
2982 char *from;
6902d925 2983 struct dentry *dentry;
dae6ad8f 2984 struct path path;
1da177e4
LT		 2985
2986 from = getname(oldname);
2ad94ae6 2987 if (IS_ERR(from))
1da177e4 2988 return PTR_ERR(from);
1da177e4 2989
dae6ad8f 2990 dentry = user_path_create(newdfd, newname, &path, 0);
6902d925
DH		 2991 error = PTR_ERR(dentry);
2992 if (IS_ERR(dentry))
dae6ad8f 2993 goto out_putname;
6902d925 2994
dae6ad8f 2995 error = mnt_want_write(path.mnt);
75c3f29d
DH		 2996 if (error)
2997 goto out_dput;
dae6ad8f 2998 error = security_path_symlink(&path, dentry, from);
be6d3e56
KT		 2999 if (error)
3000 goto out_drop_write;
dae6ad8f 3001 error = vfs_symlink(path.dentry->d_inode, dentry, from);
be6d3e56 3002out_drop_write:
dae6ad8f 3003 mnt_drop_write(path.mnt);
75c3f29d 3004out_dput:
6902d925 3005 dput(dentry);
dae6ad8f
AV		 3006 mutex_unlock(&path.dentry->d_inode->i_mutex);
3007 path_put(&path);
6902d925 3008out_putname:
1da177e4
LT		 3009 putname(from);
3010 return error;
3011}
3012
3480b257 3013SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3014{
3015 return sys_symlinkat(oldname, AT_FDCWD, newname);
3016}
3017
1da177e4
LT		 3018int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3019{
3020 struct inode *inode = old_dentry->d_inode;
8de52778 3021 unsigned max_links = dir->i_sb->s_max_links;
1da177e4
LT		 3022 int error;
3023
3024 if (!inode)
3025 return -ENOENT;
3026
a95164d9 3027 error = may_create(dir, new_dentry);
1da177e4
LT		 3028 if (error)
3029 return error;
3030
3031 if (dir->i_sb != inode->i_sb)
3032 return -EXDEV;
3033
3034 /*
3035 * A link to an append-only or immutable file cannot be created.
3036 */
3037 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3038 return -EPERM;
acfa4380 3039 if (!dir->i_op->link)
1da177e4 3040 return -EPERM;
7e79eedb 3041 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 3042 return -EPERM;
3043
3044 error = security_inode_link(old_dentry, dir, new_dentry);
3045 if (error)
3046 return error;
3047
7e79eedb 3048 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 3049 /* Make sure we don't allow creating hardlink to an unlinked file */
3050 if (inode->i_nlink == 0)
3051 error = -ENOENT;
8de52778
AV		 3052 else if (max_links && inode->i_nlink >= max_links)
3053 error = -EMLINK;
aae8a97d
AK		 3054 else
3055 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 3056 mutex_unlock(&inode->i_mutex);
e31e14ec 3057 if (!error)
7e79eedb 3058 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 3059 return error;
3060}
3061
3062/*
3063 * Hardlinks are often used in delicate situations. We avoid
3064 * security-related surprises by not following symlinks on the
3065 * newname. --KAB
3066 *
3067 * We don't follow them on the oldname either to be compatible
3068 * with linux 2.0, and to avoid hard-linking to directories
3069 * and other special files. --ADM
3070 */
2e4d0924
HC		 3071SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3072 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 3073{
3074 struct dentry *new_dentry;
dae6ad8f 3075 struct path old_path, new_path;
11a7b371 3076 int how = 0;
1da177e4 3077 int error;
1da177e4 3078
11a7b371 3079 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 3080 return -EINVAL;
11a7b371
AK		 3081 /*
3082 * To use null names we require CAP_DAC_READ_SEARCH
3083 * This ensures that not everyone will be able to create
3084 * handlink using the passed filedescriptor.
3085 */
3086 if (flags & AT_EMPTY_PATH) {
3087 if (!capable(CAP_DAC_READ_SEARCH))
3088 return -ENOENT;
3089 how = LOOKUP_EMPTY;
3090 }
3091
3092 if (flags & AT_SYMLINK_FOLLOW)
3093 how |= LOOKUP_FOLLOW;
c04030e1 3094
11a7b371 3095 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 3096 if (error)
2ad94ae6
AV		 3097 return error;
3098
dae6ad8f 3099 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
1da177e4 3100 error = PTR_ERR(new_dentry);
6902d925 3101 if (IS_ERR(new_dentry))
dae6ad8f
AV		 3102 goto out;
3103
3104 error = -EXDEV;
3105 if (old_path.mnt != new_path.mnt)
3106 goto out_dput;
3107 error = mnt_want_write(new_path.mnt);
75c3f29d
DH		 3108 if (error)
3109 goto out_dput;
dae6ad8f 3110 error = security_path_link(old_path.dentry, &new_path, new_dentry);
be6d3e56
KT		 3111 if (error)
3112 goto out_drop_write;
dae6ad8f 3113 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
be6d3e56 3114out_drop_write:
dae6ad8f 3115 mnt_drop_write(new_path.mnt);
75c3f29d 3116out_dput:
6902d925 3117 dput(new_dentry);
dae6ad8f
AV		 3118 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3119 path_put(&new_path);
1da177e4 3120out:
2d8f3038 3121 path_put(&old_path);
1da177e4
LT		 3122
3123 return error;
3124}
3125
3480b257 3126SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 3127{
c04030e1 3128 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 3129}
3130
1da177e4
LT		 3131/*
3132 * The worst of all namespace operations - renaming directory. "Perverted"
3133 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3134 * Problems:
3135 * a) we can get into loop creation. Check is done in is_subdir().
3136 * b) race potential - two innocent renames can create a loop together.
3137 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 3138 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 3139 * story.
3140 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 3141 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 3142 * whether the target exists). Solution: try to be smart with locking
3143 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 3144 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 3145 * move will be locked. Thus we can rank directories by the tree
3146 * (ancestors first) and rank all non-directories after them.
3147 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 3148 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 3149 * HOWEVER, it relies on the assumption that any object with ->lookup()
3150 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3151 * we'd better make sure that there's no link(2) for them.
e4eaac06 3152 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 3153 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 3154 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 3155 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 3156 * locking].
3157 */
75c96f85
AB		 3158static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3159 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 3160{
3161 int error = 0;
9055cba7 3162 struct inode *target = new_dentry->d_inode;
8de52778 3163 unsigned max_links = new_dir->i_sb->s_max_links;
1da177e4
LT		 3164
3165 /*
3166 * If we are going to change the parent - check write permissions,
3167 * we'll need to flip '..'.
3168 */
3169 if (new_dir != old_dir) {
f419a2e3 3170 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 3171 if (error)
3172 return error;
3173 }
3174
3175 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3176 if (error)
3177 return error;
3178
1d2ef590 3179 dget(new_dentry);
d83c49f3 3180 if (target)
1b1dcc1b 3181 mutex_lock(&target->i_mutex);
9055cba7
SW		 3182
3183 error = -EBUSY;
3184 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3185 goto out;
3186
8de52778
AV		 3187 error = -EMLINK;
3188 if (max_links && !target && new_dir != old_dir &&
3189 new_dir->i_nlink >= max_links)
3190 goto out;
3191
3cebde24
SW		 3192 if (target)
3193 shrink_dcache_parent(new_dentry);
9055cba7
SW		 3194 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3195 if (error)
3196 goto out;
3197
1da177e4 3198 if (target) {
9055cba7
SW		 3199 target->i_flags |= S_DEAD;
3200 dont_mount(new_dentry);
1da177e4 3201 }
9055cba7
SW		 3202out:
3203 if (target)
3204 mutex_unlock(&target->i_mutex);
1d2ef590 3205 dput(new_dentry);
e31e14ec 3206 if (!error)
349457cc
MF		 3207 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3208 d_move(old_dentry,new_dentry);
1da177e4
LT		 3209 return error;
3210}
3211
75c96f85
AB		 3212static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3213 struct inode *new_dir, struct dentry *new_dentry)
1da177e4 3214{
51892bbb 3215 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3216 int error;
3217
3218 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3219 if (error)
3220 return error;
3221
3222 dget(new_dentry);
1da177e4 3223 if (target)
1b1dcc1b 3224 mutex_lock(&target->i_mutex);
51892bbb
SW		 3225
3226 error = -EBUSY;
1da177e4 3227 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
51892bbb
SW		 3228 goto out;
3229
3230 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3231 if (error)
3232 goto out;
3233
3234 if (target)
3235 dont_mount(new_dentry);
3236 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3237 d_move(old_dentry, new_dentry);
3238out:
1da177e4 3239 if (target)
1b1dcc1b 3240 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3241 dput(new_dentry);
3242 return error;
3243}
3244
3245int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3246 struct inode *new_dir, struct dentry *new_dentry)
3247{
3248 int error;
3249 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3250 const unsigned char *old_name;
1da177e4
LT		 3251
3252 if (old_dentry->d_inode == new_dentry->d_inode)
3253 return 0;
3254
3255 error = may_delete(old_dir, old_dentry, is_dir);
3256 if (error)
3257 return error;
3258
3259 if (!new_dentry->d_inode)
a95164d9 3260 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3261 else
3262 error = may_delete(new_dir, new_dentry, is_dir);
3263 if (error)
3264 return error;
3265
acfa4380 3266 if (!old_dir->i_op->rename)
1da177e4
LT		 3267 return -EPERM;
3268
0eeca283
RL		 3269 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3270
1da177e4
LT		 3271 if (is_dir)
3272 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3273 else
3274 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3275 if (!error)
3276 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3277 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3278 fsnotify_oldname_free(old_name);
3279
1da177e4
LT		 3280 return error;
3281}
3282
2e4d0924
HC		 3283SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3284 int, newdfd, const char __user *, newname)
1da177e4 3285{
2ad94ae6
AV		 3286 struct dentry *old_dir, *new_dir;
3287 struct dentry *old_dentry, *new_dentry;
3288 struct dentry *trap;
1da177e4 3289 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3290 char *from;
3291 char *to;
3292 int error;
1da177e4 3293
2ad94ae6 3294 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3295 if (error)
3296 goto exit;
3297
2ad94ae6 3298 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3299 if (error)
3300 goto exit1;
3301
3302 error = -EXDEV;
4ac91378 3303 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3304 goto exit2;
3305
4ac91378 3306 old_dir = oldnd.path.dentry;
1da177e4
LT		 3307 error = -EBUSY;
3308 if (oldnd.last_type != LAST_NORM)
3309 goto exit2;
3310
4ac91378 3311 new_dir = newnd.path.dentry;
1da177e4
LT		 3312 if (newnd.last_type != LAST_NORM)
3313 goto exit2;
3314
0612d9fb
OH		 3315 oldnd.flags &= ~LOOKUP_PARENT;
3316 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3317 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3318
1da177e4
LT		 3319 trap = lock_rename(new_dir, old_dir);
3320
49705b77 3321 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3322 error = PTR_ERR(old_dentry);
3323 if (IS_ERR(old_dentry))
3324 goto exit3;
3325 /* source must exist */
3326 error = -ENOENT;
3327 if (!old_dentry->d_inode)
3328 goto exit4;
3329 /* unless the source is a directory trailing slashes give -ENOTDIR */
3330 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3331 error = -ENOTDIR;
3332 if (oldnd.last.name[oldnd.last.len])
3333 goto exit4;
3334 if (newnd.last.name[newnd.last.len])
3335 goto exit4;
3336 }
3337 /* source should not be ancestor of target */
3338 error = -EINVAL;
3339 if (old_dentry == trap)
3340 goto exit4;
49705b77 3341 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3342 error = PTR_ERR(new_dentry);
3343 if (IS_ERR(new_dentry))
3344 goto exit4;
3345 /* target should not be an ancestor of source */
3346 error = -ENOTEMPTY;
3347 if (new_dentry == trap)
3348 goto exit5;
3349
9079b1eb
DH		 3350 error = mnt_want_write(oldnd.path.mnt);
3351 if (error)
3352 goto exit5;
be6d3e56
KT		 3353 error = security_path_rename(&oldnd.path, old_dentry,
3354 &newnd.path, new_dentry);
3355 if (error)
3356 goto exit6;
1da177e4
LT		 3357 error = vfs_rename(old_dir->d_inode, old_dentry,
3358 new_dir->d_inode, new_dentry);
be6d3e56 3359exit6:
9079b1eb 3360 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3361exit5:
3362 dput(new_dentry);
3363exit4:
3364 dput(old_dentry);
3365exit3:
3366 unlock_rename(new_dir, old_dir);
3367exit2:
1d957f9b 3368 path_put(&newnd.path);
2ad94ae6 3369 putname(to);
1da177e4 3370exit1:
1d957f9b 3371 path_put(&oldnd.path);
1da177e4 3372 putname(from);
2ad94ae6 3373exit:
1da177e4
LT		 3374 return error;
3375}
3376
a26eab24 3377SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3378{
3379 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3380}
3381
1da177e4
LT		 3382int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3383{
3384 int len;
3385
3386 len = PTR_ERR(link);
3387 if (IS_ERR(link))
3388 goto out;
3389
3390 len = strlen(link);
3391 if (len > (unsigned) buflen)
3392 len = buflen;
3393 if (copy_to_user(buffer, link, len))
3394 len = -EFAULT;
3395out:
3396 return len;
3397}
3398
3399/*
3400 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3401 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3402 * using) it for any given inode is up to filesystem.
3403 */
3404int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3405{
3406 struct nameidata nd;
cc314eef 3407 void *cookie;
694a1764 3408 int res;
cc314eef 3409
1da177e4 3410 nd.depth = 0;
cc314eef 3411 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3412 if (IS_ERR(cookie))
3413 return PTR_ERR(cookie);
3414
3415 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3416 if (dentry->d_inode->i_op->put_link)
3417 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3418 return res;
1da177e4
LT		 3419}
3420
3421int vfs_follow_link(struct nameidata *nd, const char *link)
3422{
3423 return __vfs_follow_link(nd, link);
3424}
3425
3426/* get the link contents into pagecache */
3427static char *page_getlink(struct dentry * dentry, struct page **ppage)
3428{
ebd09abb
DG		 3429 char *kaddr;
3430 struct page *page;
1da177e4 3431 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3432 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3433 if (IS_ERR(page))
6fe6900e 3434 return (char*)page;
1da177e4 3435 *ppage = page;
ebd09abb
DG		 3436 kaddr = kmap(page);
3437 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3438 return kaddr;
1da177e4
LT		 3439}
3440
3441int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3442{
3443 struct page *page = NULL;
3444 char *s = page_getlink(dentry, &page);
3445 int res = vfs_readlink(dentry,buffer,buflen,s);
3446 if (page) {
3447 kunmap(page);
3448 page_cache_release(page);
3449 }
3450 return res;
3451}
3452
cc314eef 3453void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3454{
cc314eef 3455 struct page *page = NULL;
1da177e4 3456 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3457 return page;
1da177e4
LT		 3458}
3459
cc314eef 3460void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3461{
cc314eef
LT		 3462 struct page *page = cookie;
3463
3464 if (page) {
1da177e4
LT		 3465 kunmap(page);
3466 page_cache_release(page);
1da177e4
LT		 3467 }
3468}
3469
54566b2c
NP		 3470/*
3471 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3472 */
3473int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3474{
3475 struct address_space *mapping = inode->i_mapping;
0adb25d2 3476 struct page *page;
afddba49 3477 void *fsdata;
beb497ab 3478 int err;
1da177e4 3479 char *kaddr;
54566b2c
NP		 3480 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3481 if (nofs)
3482 flags |= AOP_FLAG_NOFS;
1da177e4 3483
7e53cac4 3484retry:
afddba49 3485 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3486 flags, &page, &fsdata);
1da177e4 3487 if (err)
afddba49
NP		 3488 goto fail;
3489
e8e3c3d6 3490 kaddr = kmap_atomic(page);
1da177e4 3491 memcpy(kaddr, symname, len-1);
e8e3c3d6 3492 kunmap_atomic(kaddr);
afddba49
NP		 3493
3494 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3495 page, fsdata);
1da177e4
LT		 3496 if (err < 0)
3497 goto fail;
afddba49
NP		 3498 if (err < len-1)
3499 goto retry;
3500
1da177e4
LT		 3501 mark_inode_dirty(inode);
3502 return 0;
1da177e4
LT		 3503fail:
3504 return err;
3505}
3506
0adb25d2
KK		 3507int page_symlink(struct inode *inode, const char *symname, int len)
3508{
3509 return __page_symlink(inode, symname, len,
54566b2c 3510 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3511}
3512
92e1d5be 3513const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3514 .readlink = generic_readlink,
3515 .follow_link = page_follow_link_light,
3516 .put_link = page_put_link,
3517};
3518
2d8f3038 3519EXPORT_SYMBOL(user_path_at);
cc53ce53 3520EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3521EXPORT_SYMBOL(follow_down);
3522EXPORT_SYMBOL(follow_up);
3523EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3524EXPORT_SYMBOL(getname);
3525EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3526EXPORT_SYMBOL(lookup_one_len);
3527EXPORT_SYMBOL(page_follow_link_light);
3528EXPORT_SYMBOL(page_put_link);
3529EXPORT_SYMBOL(page_readlink);
0adb25d2 3530EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3531EXPORT_SYMBOL(page_symlink);
3532EXPORT_SYMBOL(page_symlink_inode_operations);
d1811465 3533EXPORT_SYMBOL(kern_path);
16f18200 3534EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3535EXPORT_SYMBOL(inode_permission);
1da177e4
LT		 3536EXPORT_SYMBOL(unlock_rename);
3537EXPORT_SYMBOL(vfs_create);
3538EXPORT_SYMBOL(vfs_follow_link);
3539EXPORT_SYMBOL(vfs_link);
3540EXPORT_SYMBOL(vfs_mkdir);
3541EXPORT_SYMBOL(vfs_mknod);
3542EXPORT_SYMBOL(generic_permission);
3543EXPORT_SYMBOL(vfs_readlink);
3544EXPORT_SYMBOL(vfs_rename);
3545EXPORT_SYMBOL(vfs_rmdir);
3546EXPORT_SYMBOL(vfs_symlink);
3547EXPORT_SYMBOL(vfs_unlink);
3548EXPORT_SYMBOL(dentry_unhash);
3549EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)